applied-ai-018 commited on
Commit
dffa97c
·
verified ·
1 Parent(s): 64dbae1

Add files using upload-large-folder tool

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. ckpts/universal/global_step20/zero/10.mlp.dense_h_to_4h.weight/exp_avg.pt +3 -0
  2. ckpts/universal/global_step20/zero/10.mlp.dense_h_to_4h.weight/fp32.pt +3 -0
  3. ckpts/universal/global_step20/zero/14.post_attention_layernorm.weight/exp_avg_sq.pt +3 -0
  4. ckpts/universal/global_step20/zero/6.post_attention_layernorm.weight/exp_avg.pt +3 -0
  5. ckpts/universal/global_step20/zero/6.post_attention_layernorm.weight/exp_avg_sq.pt +3 -0
  6. ckpts/universal/global_step20/zero/9.mlp.dense_h_to_4h_swiglu.weight/exp_avg_sq.pt +3 -0
  7. lm-evaluation-harness/lm_eval/tasks/agieval/aqua-rat.yaml +24 -0
  8. lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-biology.yaml +6 -0
  9. lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-chemistry.yaml +6 -0
  10. lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-chinese.yaml +6 -0
  11. lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-english.yaml +6 -0
  12. lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-history.yaml +6 -0
  13. lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-mathqa.yaml +6 -0
  14. lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-physics.yaml +6 -0
  15. lm-evaluation-harness/lm_eval/tasks/agieval/jec-qa-ca.yaml +6 -0
  16. lm-evaluation-harness/lm_eval/tasks/agieval/jec-qa-kd.yaml +6 -0
  17. lm-evaluation-harness/lm_eval/tasks/agieval/logiqa-en.yaml +7 -0
  18. lm-evaluation-harness/lm_eval/tasks/agieval/lsat-ar.yaml +7 -0
  19. lm-evaluation-harness/lm_eval/tasks/agieval/lsat-rc.yaml +7 -0
  20. lm-evaluation-harness/lm_eval/tasks/agieval/sat-en-without-passage.yaml +7 -0
  21. lm-evaluation-harness/lm_eval/tasks/agieval/sat-math.yaml +7 -0
  22. lm-evaluation-harness/lm_eval/tasks/agieval/utils.py +274 -0
  23. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_as.yaml +33 -0
  24. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_gu.yaml +33 -0
  25. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_kn.yaml +33 -0
  26. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_ml.yaml +33 -0
  27. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_ne.yaml +33 -0
  28. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_sa.yaml +33 -0
  29. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_sd.yaml +33 -0
  30. lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_ur.yaml +33 -0
  31. lm-evaluation-harness/lm_eval/tasks/indiccopa/utils.py +136 -0
  32. venv/lib/python3.10/site-packages/datasets/combine.py +215 -0
  33. venv/lib/python3.10/site-packages/datasets/dataset_dict.py +0 -0
  34. venv/lib/python3.10/site-packages/datasets/exceptions.py +85 -0
  35. venv/lib/python3.10/site-packages/datasets/inspect.py +582 -0
  36. venv/lib/python3.10/site-packages/datasets/streaming.py +142 -0
  37. venv/lib/python3.10/site-packages/datasets/table.py +2415 -0
  38. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Bermuda +0 -0
  39. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Canary +0 -0
  40. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Cape_Verde +0 -0
  41. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Faeroe +0 -0
  42. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Faroe +0 -0
  43. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Jan_Mayen +0 -0
  44. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Reykjavik +0 -0
  45. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/South_Georgia +0 -0
  46. venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Stanley +0 -0
  47. venv/lib/python3.10/site-packages/pytz/zoneinfo/Mexico/BajaNorte +0 -0
  48. venv/lib/python3.10/site-packages/pytz/zoneinfo/Mexico/BajaSur +0 -0
  49. venv/lib/python3.10/site-packages/pytz/zoneinfo/Mexico/General +0 -0
  50. venv/lib/python3.10/site-packages/pytz/zoneinfo/Pacific/Apia +0 -0
ckpts/universal/global_step20/zero/10.mlp.dense_h_to_4h.weight/exp_avg.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d64b94bf893224ff1789b692a88a7052307ab1a947134c2713501d2b119513fd
3
+ size 33555612
ckpts/universal/global_step20/zero/10.mlp.dense_h_to_4h.weight/fp32.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d5ba02121cdc9ced8147ef6bc5084f2ea1e7e9ddcb3bb75514896fcdf78d0dd1
3
+ size 33555533
ckpts/universal/global_step20/zero/14.post_attention_layernorm.weight/exp_avg_sq.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cb5380630c6eb51214baa0c22f53c8663bdabbeee5ec91aa1bddf01b366fc063
3
+ size 9387
ckpts/universal/global_step20/zero/6.post_attention_layernorm.weight/exp_avg.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d7f91ef94428e9dcc46ab62e13fa34130086a40dfa1388cb432205029a078fc0
3
+ size 9372
ckpts/universal/global_step20/zero/6.post_attention_layernorm.weight/exp_avg_sq.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8dcb79d73925597af47f3730c3fbebbab533108fb86d4e7741e572d1699a7ae
3
+ size 9387
ckpts/universal/global_step20/zero/9.mlp.dense_h_to_4h_swiglu.weight/exp_avg_sq.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f97da87e4a3cd5212cf743dd0ade277cf4c1c6847eb0415ec868ebe426b46dc3
3
+ size 33555627
lm-evaluation-harness/lm_eval/tasks/agieval/aqua-rat.yaml ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ group:
2
+ - agieval
3
+ - agieval_en
4
+ - agieval_nous
5
+ task: agieval_aqua_rat
6
+ dataset_path: hails/agieval-aqua-rat
7
+ dataset_name: null
8
+ output_type: multiple_choice
9
+ training_split: null
10
+ validation_split: null
11
+ test_split: test
12
+ doc_to_text: "{{query}}"
13
+ doc_to_target: "{{gold}}"
14
+ doc_to_choice: "{{choices}}"
15
+ process_results: !function utils.process_results_mcqa
16
+ metric_list:
17
+ - metric: acc
18
+ aggregation: mean
19
+ higher_is_better: true
20
+ - metric: acc_norm
21
+ aggregation: mean
22
+ higher_is_better: true
23
+ metadata:
24
+ version: 1.0
lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-biology.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_gaokao_biology
6
+ dataset_path: hails/agieval-gaokao-biology
lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-chemistry.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_gaokao_chemistry
6
+ dataset_path: hails/agieval-gaokao-chemistry
lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-chinese.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_gaokao_chinese
6
+ dataset_path: hails/agieval-gaokao-chinese
lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-english.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_en # categorizing as EN because the AGIEval codebase lists this as in `english_qa_tasks`
5
+ task: agieval_gaokao_english
6
+ dataset_path: hails/agieval-gaokao-english
lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-history.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_gaokao_history
6
+ dataset_path: hails/agieval-gaokao-history
lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-mathqa.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_gaokao_mathqa
6
+ dataset_path: hails/agieval-gaokao-mathqa
lm-evaluation-harness/lm_eval/tasks/agieval/gaokao-physics.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_gaokao_physics
6
+ dataset_path: hails/agieval-gaokao-physics
lm-evaluation-harness/lm_eval/tasks/agieval/jec-qa-ca.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_jec_qa_ca
6
+ dataset_path: hails/agieval-jec-qa-ca
lm-evaluation-harness/lm_eval/tasks/agieval/jec-qa-kd.yaml ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_cn
5
+ task: agieval_jec_qa_kd
6
+ dataset_path: hails/agieval-jec-qa-kd
lm-evaluation-harness/lm_eval/tasks/agieval/logiqa-en.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_nous
5
+ - agieval_en
6
+ task: agieval_logiqa_en
7
+ dataset_path: hails/agieval-logiqa-en
lm-evaluation-harness/lm_eval/tasks/agieval/lsat-ar.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_nous
5
+ - agieval_en
6
+ task: agieval_lsat_ar
7
+ dataset_path: hails/agieval-lsat-ar
lm-evaluation-harness/lm_eval/tasks/agieval/lsat-rc.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_nous
5
+ - agieval_en
6
+ task: agieval_lsat_rc
7
+ dataset_path: hails/agieval-lsat-rc
lm-evaluation-harness/lm_eval/tasks/agieval/sat-en-without-passage.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_nous
5
+ - agieval_en
6
+ task: agieval_sat_en_without_passage
7
+ dataset_path: hails/agieval-sat-en-without-passage
lm-evaluation-harness/lm_eval/tasks/agieval/sat-math.yaml ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ include: aqua-rat.yaml
2
+ group:
3
+ - agieval
4
+ - agieval_nous
5
+ - agieval_en
6
+ task: agieval_sat_math
7
+ dataset_path: hails/agieval-sat-math
lm-evaluation-harness/lm_eval/tasks/agieval/utils.py ADDED
@@ -0,0 +1,274 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Answer parsing and normalization code, from
2
+ # https://github.com/ruixiangcui/AGIEval/blob/main/src/
3
+ # math_equivalence.py and post_process.py
4
+ import re
5
+ from typing import Dict, List
6
+
7
+ import numpy as np
8
+
9
+
10
+ def parse_math_answer(raw_string):
11
+ def remove_boxed(s):
12
+ left = "\\boxed{"
13
+ try:
14
+ assert s[: len(left)] == left
15
+ assert s[-1] == "}"
16
+ answer = s[len(left) : -1]
17
+ if "=" in answer:
18
+ answer = answer.split("=")[-1].lstrip(" ")
19
+ return answer
20
+ except Exception:
21
+ return None
22
+
23
+ def last_boxed_only_string(string):
24
+ idx = string.rfind("\\boxed")
25
+ if idx < 0:
26
+ idx = string.rfind("\\fbox")
27
+ if idx < 0:
28
+ return None
29
+ i = idx
30
+ right_brace_idx = None
31
+ num_left_braces_open = 0
32
+ while i < len(string):
33
+ if string[i] == "{":
34
+ num_left_braces_open += 1
35
+ if string[i] == "}":
36
+ num_left_braces_open -= 1
37
+ if num_left_braces_open == 0:
38
+ right_brace_idx = i
39
+ break
40
+ i += 1
41
+
42
+ if right_brace_idx is None:
43
+ retval = None
44
+ else:
45
+ retval = string[idx : right_brace_idx + 1]
46
+
47
+ return retval
48
+
49
+ def get_answer_with_dollar_sign(s):
50
+ first_pattern = "\$(.*)\$"
51
+ last_match = None
52
+ matches = re.findall(first_pattern, s)
53
+ if matches:
54
+ last_match = matches[-1]
55
+ if "=" in last_match:
56
+ last_match = last_match.split("=")[-1].lstrip(" ")
57
+ return last_match
58
+
59
+ def get_answer_without_dollar_sign(s):
60
+ last_match = None
61
+ if "=" in s:
62
+ last_match = s.split("=")[-1].lstrip(" ").rstrip(".")
63
+ if "\\n" in last_match:
64
+ last_match = last_match.split("\\n")[0]
65
+ else:
66
+ pattern = "(?:\\$)?\d+(?:\.\d+)?(?![\w\d])"
67
+ matches = re.findall(pattern, s)
68
+ if matches:
69
+ last_match = matches[-1]
70
+ return last_match
71
+
72
+ if "\\boxed" in raw_string:
73
+ answer = remove_boxed(last_boxed_only_string(raw_string))
74
+ else:
75
+ answer = get_answer_with_dollar_sign(raw_string)
76
+ if not answer:
77
+ answer = get_answer_without_dollar_sign(raw_string)
78
+ return answer
79
+
80
+
81
+ # code from https://github.com/hendrycks/math/blob/main/modeling/math_equivalence.py
82
+ def _fix_fracs(string):
83
+ substrs = string.split("\\frac")
84
+ new_str = substrs[0]
85
+ if len(substrs) > 1:
86
+ substrs = substrs[1:]
87
+ for substr in substrs:
88
+ new_str += "\\frac"
89
+ if substr[0] == "{":
90
+ new_str += substr
91
+ else:
92
+ try:
93
+ assert len(substr) >= 2
94
+ except Exception:
95
+ return string
96
+ a = substr[0]
97
+ b = substr[1]
98
+ if b != "{":
99
+ if len(substr) > 2:
100
+ post_substr = substr[2:]
101
+ new_str += "{" + a + "}{" + b + "}" + post_substr
102
+ else:
103
+ new_str += "{" + a + "}{" + b + "}"
104
+ else:
105
+ if len(substr) > 2:
106
+ post_substr = substr[2:]
107
+ new_str += "{" + a + "}" + b + post_substr
108
+ else:
109
+ new_str += "{" + a + "}" + b
110
+ string = new_str
111
+ return string
112
+
113
+
114
+ def _fix_a_slash_b(string):
115
+ if len(string.split("/")) != 2:
116
+ return string
117
+ a = string.split("/")[0]
118
+ b = string.split("/")[1]
119
+ try:
120
+ a = int(a)
121
+ b = int(b)
122
+ assert string == "{}/{}".format(a, b)
123
+ new_string = "\\frac{" + str(a) + "}{" + str(b) + "}"
124
+ return new_string
125
+ except Exception:
126
+ return string
127
+
128
+
129
+ def _remove_right_units(string):
130
+ # "\\text{ " only ever occurs (at least in the val set) when describing units
131
+ if "\\text{ " in string:
132
+ splits = string.split("\\text{ ")
133
+ assert len(splits) == 2
134
+ return splits[0]
135
+ else:
136
+ return string
137
+
138
+
139
+ def _fix_sqrt(string):
140
+ if "\\sqrt" not in string:
141
+ return string
142
+ splits = string.split("\\sqrt")
143
+ new_string = splits[0]
144
+ for split in splits[1:]:
145
+ if split[0] != "{":
146
+ a = split[0]
147
+ new_substr = "\\sqrt{" + a + "}" + split[1:]
148
+ else:
149
+ new_substr = "\\sqrt" + split
150
+ new_string += new_substr
151
+ return new_string
152
+
153
+
154
+ def _strip_string(string):
155
+ # linebreaks
156
+ string = string.replace("\n", "")
157
+ # print(string)
158
+
159
+ # remove inverse spaces
160
+ string = string.replace("\\!", "")
161
+ # print(string)
162
+
163
+ # replace \\ with \
164
+ string = string.replace("\\\\", "\\")
165
+ # print(string)
166
+
167
+ # replace tfrac and dfrac with frac
168
+ string = string.replace("tfrac", "frac")
169
+ string = string.replace("dfrac", "frac")
170
+ # print(string)
171
+
172
+ # remove \left and \right
173
+ string = string.replace("\\left", "")
174
+ string = string.replace("\\right", "")
175
+ # print(string)
176
+
177
+ # Remove circ (degrees)
178
+ string = string.replace("^{\\circ}", "")
179
+ string = string.replace("^\\circ", "")
180
+
181
+ # remove dollar signs
182
+ string = string.replace("\\$", "")
183
+
184
+ # remove units (on the right)
185
+ string = _remove_right_units(string)
186
+
187
+ # remove percentage
188
+ string = string.replace("\\%", "")
189
+ string = string.replace("\%", "")
190
+
191
+ # " 0." equivalent to " ." and "{0." equivalent to "{." Alternatively, add "0" if "." is the start of the string
192
+ string = string.replace(" .", " 0.")
193
+ string = string.replace("{.", "{0.")
194
+ # if empty, return empty string
195
+ if len(string) == 0:
196
+ return string
197
+ if string[0] == ".":
198
+ string = "0" + string
199
+
200
+ # to consider: get rid of e.g. "k = " or "q = " at beginning
201
+ if len(string.split("=")) == 2:
202
+ if len(string.split("=")[0]) <= 2:
203
+ string = string.split("=")[1]
204
+
205
+ # fix sqrt3 --> sqrt{3}
206
+ string = _fix_sqrt(string)
207
+
208
+ # remove spaces
209
+ string = string.replace(" ", "")
210
+
211
+ # \frac1b or \frac12 --> \frac{1}{b} and \frac{1}{2}, etc. Even works with \frac1{72} (but not \frac{72}1). Also does a/b --> \\frac{a}{b}
212
+ string = _fix_fracs(string)
213
+
214
+ # manually change 0.5 --> \frac{1}{2}
215
+ if string == "0.5":
216
+ string = "\\frac{1}{2}"
217
+
218
+ # NOTE: X/Y changed to \frac{X}{Y} in dataset, but in simple cases fix in case the model output is X/Y
219
+ string = _fix_a_slash_b(string)
220
+
221
+ return string
222
+
223
+
224
+ def is_equiv(str1, str2, verbose=False):
225
+ if str1 is None and str2 is None:
226
+ print("WARNING: Both None")
227
+ return True
228
+ if str1 is None or str2 is None:
229
+ return False
230
+
231
+ str1, str2 = parse_math_answer(str1), parse_math_answer(str2)
232
+
233
+ try:
234
+ ss1 = _strip_string(str1)
235
+ ss2 = _strip_string(str2)
236
+ if verbose:
237
+ print(ss1, ss2)
238
+ return ss1 == ss2
239
+ except Exception:
240
+ return str1 == str2
241
+
242
+
243
+ def process_results(doc: dict, results: List[str]) -> Dict[str, int]:
244
+ candidate = results[0]
245
+
246
+ gold = doc["answer"]
247
+
248
+ if not gold:
249
+ print(doc, candidate, gold)
250
+ if is_equiv(candidate, gold):
251
+ retval = 1
252
+ else:
253
+ retval = 0
254
+
255
+ results = {
256
+ "acc": retval,
257
+ }
258
+ return results
259
+
260
+
261
+ # use a custom process_results() function, because AGIEval can have multiple valid answers
262
+ def process_results_mcqa(doc, results):
263
+ results = [result[0] for result in results]
264
+
265
+ gold = doc["gold"]
266
+
267
+ acc = 1.0 if int(np.argmax(results)) in gold else 0.0
268
+ completion_len = np.array([float(len(i)) for i in doc["choices"]])
269
+ acc_norm = 1.0 if int(np.argmax(results / completion_len)) in gold else 0.0
270
+
271
+ return {
272
+ "acc": acc,
273
+ "acc_norm": acc_norm,
274
+ }
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_as.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tass file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-as
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # asgher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_as
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-as
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_gu.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tgus file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-gu
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # gugher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_gu
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-gu
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_kn.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tkns file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-kn
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # kngher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_kn
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-kn
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_ml.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tmls file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-ml
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # mlgher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_ml
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-ml
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_ne.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tnes file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-ne
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # negher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_ne
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-ne
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_sa.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tsas file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-sa
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # sagher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_sa
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-sa
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_sd.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Tsds file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-sd
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # sdgher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_sd
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-sd
lm-evaluation-harness/lm_eval/tasks/indiccopa/indiccopa_ur.yaml ADDED
@@ -0,0 +1,33 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Turs file will be included in the generated language-specific task configs.
2
+ # It doesn't have a yaml file extension as it is not meant to be imported directly
3
+ # by the harness.
4
+ group: ai4bharat/IndicCOPA
5
+ dataset_path: ai4bharat/IndicCOPA
6
+ dataset_name: translation-ur
7
+ output_type: multiple_choice
8
+ # training_split: train
9
+ # validation_split: validation
10
+ test_split: test
11
+ # doc_to_text: "Premise: {{premise}}\nGiven the premise what is the {{question}}\nPlease Choose Among following 2 choices and label them as 0 for 1st choice and 1 for 2nd choice."
12
+ # doc_to_target: label
13
+ # doc_to_choice: "{{choice1}}{{choice2}}"
14
+ # metric_list:
15
+ # - metric: acc
16
+ # aggregation: mean
17
+ # urgher_is_better: true
18
+ # metadata:
19
+ # version: 1.0
20
+
21
+ doc_to_text: !function utils.doc_to_text_ur
22
+ doc_to_target: label
23
+ doc_to_choice: !function utils.doc_to_choice
24
+ metric_list:
25
+ - metric: acc
26
+ metadata:
27
+ version: 1.0
28
+
29
+
30
+ # doc_to_choice: '{{[premise+", सही? हाँ, "+hypothesis,premise+", सही? इसलिए, "+hypothesis,premise+",
31
+ # सही? नहीं, "+hypothesis]}}'
32
+ # doc_to_text: ''
33
+ task: indiccopa-ur
lm-evaluation-harness/lm_eval/tasks/indiccopa/utils.py ADDED
@@ -0,0 +1,136 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from functools import partial
2
+
3
+
4
+ def convert_choice(choice):
5
+ return choice
6
+
7
+
8
+ def doc_to_text(doc, connector):
9
+ # Drop the period
10
+ conn = connector[doc["question"]]
11
+ return doc["premise"].strip()[:-1] + f" {conn}"
12
+
13
+
14
+ def doc_to_choice(doc):
15
+ return [convert_choice(doc["choice1"]), convert_choice(doc["choice2"])]
16
+
17
+
18
+ doc_to_text_hi = partial(
19
+ doc_to_text,
20
+ connector={
21
+ "cause": "कारण",
22
+ "effect": "परिणाम",
23
+ },
24
+ )
25
+
26
+ doc_to_text_mr = partial(
27
+ doc_to_text,
28
+ connector={
29
+ "cause": "कारण",
30
+ "effect": "परिणाम",
31
+ },
32
+ )
33
+
34
+ doc_to_text_as = partial(
35
+ doc_to_text,
36
+ connector={
37
+ "cause": "কাৰণ",
38
+ "effect": "প্ৰভাৱ",
39
+ },
40
+ )
41
+
42
+ doc_to_text_bn = partial(
43
+ doc_to_text,
44
+ connector={
45
+ "cause": "কারণ",
46
+ "effect": "প্রভাব",
47
+ },
48
+ )
49
+
50
+ doc_to_text_gu = partial(
51
+ doc_to_text,
52
+ connector={
53
+ "cause": "કારણ",
54
+ "effect": "અસર",
55
+ },
56
+ )
57
+
58
+ doc_to_text_kn = partial(
59
+ doc_to_text,
60
+ connector={
61
+ "cause": "ಕಾರಣ",
62
+ "effect": "ಪರಿಣಾಮ",
63
+ },
64
+ )
65
+
66
+ doc_to_text_mai = partial(
67
+ doc_to_text,
68
+ connector={
69
+ "cause": "कारण",
70
+ "effect": "प्रभाव",
71
+ },
72
+ )
73
+
74
+ doc_to_text_ml = partial(
75
+ doc_to_text,
76
+ connector={
77
+ "cause": "കാരണമാകുന്നു",
78
+ "effect": "ഫലം",
79
+ },
80
+ )
81
+
82
+ doc_to_text_ne = partial(
83
+ doc_to_text,
84
+ connector={
85
+ "cause": "कारण",
86
+ "effect": "असर",
87
+ },
88
+ )
89
+
90
+ doc_to_text_or = partial(
91
+ doc_to_text,
92
+ connector={
93
+ "cause": "କାରଣ",
94
+ "effect": "ପ୍ରଭାବ",
95
+ },
96
+ )
97
+
98
+ doc_to_text_sa = partial(
99
+ doc_to_text,
100
+ connector={
101
+ "cause": "निमित्तम्‌",
102
+ "effect": "परिणाम",
103
+ },
104
+ )
105
+
106
+ doc_to_text_sd = partial(
107
+ doc_to_text,
108
+ connector={
109
+ "cause": "سبب",
110
+ "effect": "اثر",
111
+ },
112
+ )
113
+
114
+ doc_to_text_ta = partial(
115
+ doc_to_text,
116
+ connector={
117
+ "cause": "காரணம்",
118
+ "effect": "விளைவு",
119
+ },
120
+ )
121
+
122
+ doc_to_text_te = partial(
123
+ doc_to_text,
124
+ connector={
125
+ "cause": "కారణం",
126
+ "effect": "ప్రభావం",
127
+ },
128
+ )
129
+
130
+ doc_to_text_ur = partial(
131
+ doc_to_text,
132
+ connector={
133
+ "cause": "وجہ",
134
+ "effect": "اثر",
135
+ },
136
+ )
venv/lib/python3.10/site-packages/datasets/combine.py ADDED
@@ -0,0 +1,215 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from typing import List, Optional, TypeVar
2
+
3
+ from .arrow_dataset import Dataset, _concatenate_map_style_datasets, _interleave_map_style_datasets
4
+ from .dataset_dict import DatasetDict, IterableDatasetDict
5
+ from .info import DatasetInfo
6
+ from .iterable_dataset import IterableDataset, _concatenate_iterable_datasets, _interleave_iterable_datasets
7
+ from .splits import NamedSplit
8
+ from .utils import logging
9
+ from .utils.py_utils import Literal
10
+
11
+
12
+ logger = logging.get_logger(__name__)
13
+
14
+
15
+ DatasetType = TypeVar("DatasetType", Dataset, IterableDataset)
16
+
17
+
18
+ def interleave_datasets(
19
+ datasets: List[DatasetType],
20
+ probabilities: Optional[List[float]] = None,
21
+ seed: Optional[int] = None,
22
+ info: Optional[DatasetInfo] = None,
23
+ split: Optional[NamedSplit] = None,
24
+ stopping_strategy: Literal["first_exhausted", "all_exhausted"] = "first_exhausted",
25
+ ) -> DatasetType:
26
+ """
27
+ Interleave several datasets (sources) into a single dataset.
28
+ The new dataset is constructed by alternating between the sources to get the examples.
29
+
30
+ You can use this function on a list of [`Dataset`] objects, or on a list of [`IterableDataset`] objects.
31
+
32
+ - If `probabilities` is `None` (default) the new dataset is constructed by cycling between each source to get the examples.
33
+ - If `probabilities` is not `None`, the new dataset is constructed by getting examples from a random source at a time according to the provided probabilities.
34
+
35
+ The resulting dataset ends when one of the source datasets runs out of examples except when `oversampling` is `True`,
36
+ in which case, the resulting dataset ends when all datasets have ran out of examples at least one time.
37
+
38
+ Note for iterable datasets:
39
+
40
+ In a distributed setup or in PyTorch DataLoader workers, the stopping strategy is applied per process.
41
+ Therefore the "first_exhausted" strategy on an sharded iterable dataset can generate less samples in total (up to 1 missing sample per subdataset per worker).
42
+
43
+ Args:
44
+ datasets (`List[Dataset]` or `List[IterableDataset]`):
45
+ List of datasets to interleave.
46
+ probabilities (`List[float]`, *optional*, defaults to `None`):
47
+ If specified, the new dataset is constructed by sampling
48
+ examples from one source at a time according to these probabilities.
49
+ seed (`int`, *optional*, defaults to `None`):
50
+ The random seed used to choose a source for each example.
51
+ info ([`DatasetInfo`], *optional*):
52
+ Dataset information, like description, citation, etc.
53
+ <Added version="2.4.0"/>
54
+ split ([`NamedSplit`], *optional*):
55
+ Name of the dataset split.
56
+ <Added version="2.4.0"/>
57
+ stopping_strategy (`str`, defaults to `first_exhausted`):
58
+ Two strategies are proposed right now, `first_exhausted` and `all_exhausted`.
59
+ By default, `first_exhausted` is an undersampling strategy, i.e the dataset construction is stopped as soon as one dataset has ran out of samples.
60
+ If the strategy is `all_exhausted`, we use an oversampling strategy, i.e the dataset construction is stopped as soon as every samples of every dataset has been added at least once.
61
+ Note that if the strategy is `all_exhausted`, the interleaved dataset size can get enormous:
62
+ - with no probabilities, the resulting dataset will have `max_length_datasets*nb_dataset` samples.
63
+ - with given probabilities, the resulting dataset will have more samples if some datasets have really low probability of visiting.
64
+ Returns:
65
+ [`Dataset`] or [`IterableDataset`]: Return type depends on the input `datasets`
66
+ parameter. `Dataset` if the input is a list of `Dataset`, `IterableDataset` if the input is a list of
67
+ `IterableDataset`.
68
+
69
+ Example:
70
+
71
+ For regular datasets (map-style):
72
+
73
+ ```python
74
+ >>> from datasets import Dataset, interleave_datasets
75
+ >>> d1 = Dataset.from_dict({"a": [0, 1, 2]})
76
+ >>> d2 = Dataset.from_dict({"a": [10, 11, 12]})
77
+ >>> d3 = Dataset.from_dict({"a": [20, 21, 22]})
78
+ >>> dataset = interleave_datasets([d1, d2, d3], probabilities=[0.7, 0.2, 0.1], seed=42, stopping_strategy="all_exhausted")
79
+ >>> dataset["a"]
80
+ [10, 0, 11, 1, 2, 20, 12, 10, 0, 1, 2, 21, 0, 11, 1, 2, 0, 1, 12, 2, 10, 0, 22]
81
+ >>> dataset = interleave_datasets([d1, d2, d3], probabilities=[0.7, 0.2, 0.1], seed=42)
82
+ >>> dataset["a"]
83
+ [10, 0, 11, 1, 2]
84
+ >>> dataset = interleave_datasets([d1, d2, d3])
85
+ >>> dataset["a"]
86
+ [0, 10, 20, 1, 11, 21, 2, 12, 22]
87
+ >>> dataset = interleave_datasets([d1, d2, d3], stopping_strategy="all_exhausted")
88
+ >>> dataset["a"]
89
+ [0, 10, 20, 1, 11, 21, 2, 12, 22]
90
+ >>> d1 = Dataset.from_dict({"a": [0, 1, 2]})
91
+ >>> d2 = Dataset.from_dict({"a": [10, 11, 12, 13]})
92
+ >>> d3 = Dataset.from_dict({"a": [20, 21, 22, 23, 24]})
93
+ >>> dataset = interleave_datasets([d1, d2, d3])
94
+ >>> dataset["a"]
95
+ [0, 10, 20, 1, 11, 21, 2, 12, 22]
96
+ >>> dataset = interleave_datasets([d1, d2, d3], stopping_strategy="all_exhausted")
97
+ >>> dataset["a"]
98
+ [0, 10, 20, 1, 11, 21, 2, 12, 22, 0, 13, 23, 1, 10, 24]
99
+ >>> dataset = interleave_datasets([d1, d2, d3], probabilities=[0.7, 0.2, 0.1], seed=42)
100
+ >>> dataset["a"]
101
+ [10, 0, 11, 1, 2]
102
+ >>> dataset = interleave_datasets([d1, d2, d3], probabilities=[0.7, 0.2, 0.1], seed=42, stopping_strategy="all_exhausted")
103
+ >>> dataset["a"]
104
+ [10, 0, 11, 1, 2, 20, 12, 13, ..., 0, 1, 2, 0, 24]
105
+ For datasets in streaming mode (iterable):
106
+
107
+ >>> from datasets import load_dataset, interleave_datasets
108
+ >>> d1 = load_dataset("oscar", "unshuffled_deduplicated_en", split="train", streaming=True)
109
+ >>> d2 = load_dataset("oscar", "unshuffled_deduplicated_fr", split="train", streaming=True)
110
+ >>> dataset = interleave_datasets([d1, d2])
111
+ >>> iterator = iter(dataset)
112
+ >>> next(iterator)
113
+ {'text': 'Mtendere Village was inspired by the vision...}
114
+ >>> next(iterator)
115
+ {'text': "Média de débat d'idées, de culture...}
116
+ ```
117
+ """
118
+ from .arrow_dataset import Dataset
119
+ from .iterable_dataset import IterableDataset
120
+
121
+ if not datasets:
122
+ raise ValueError("Unable to interleave an empty list of datasets.")
123
+ for i, dataset in enumerate(datasets):
124
+ if not isinstance(dataset, (Dataset, IterableDataset)):
125
+ if isinstance(dataset, (DatasetDict, IterableDatasetDict)):
126
+ if not dataset:
127
+ raise ValueError(
128
+ f"Expected a list of Dataset objects or a list of IterableDataset objects, but element at position {i} "
129
+ "is an empty dataset dictionary."
130
+ )
131
+ raise ValueError(
132
+ f"Dataset at position {i} has at least one split: {list(dataset)}\n"
133
+ f"Please pick one to interleave with the other datasets, for example: dataset['{next(iter(dataset))}']"
134
+ )
135
+ raise ValueError(
136
+ f"Expected a list of Dataset objects or a list of IterableDataset objects, but element at position {i} is a {type(dataset).__name__}."
137
+ )
138
+ if i == 0:
139
+ dataset_type, other_type = (
140
+ (Dataset, IterableDataset) if isinstance(dataset, Dataset) else (IterableDataset, Dataset)
141
+ )
142
+ elif not isinstance(dataset, dataset_type):
143
+ raise ValueError(
144
+ f"Unable to interleave a {dataset_type.__name__} (at position 0) with a {other_type.__name__} (at position {i}). Expected a list of Dataset objects or a list of IterableDataset objects."
145
+ )
146
+ if stopping_strategy not in ["first_exhausted", "all_exhausted"]:
147
+ raise ValueError(f"{stopping_strategy} is not supported. Please enter a valid stopping_strategy.")
148
+ if dataset_type is Dataset:
149
+ return _interleave_map_style_datasets(
150
+ datasets, probabilities, seed, info=info, split=split, stopping_strategy=stopping_strategy
151
+ )
152
+ else:
153
+ return _interleave_iterable_datasets(
154
+ datasets, probabilities, seed, info=info, split=split, stopping_strategy=stopping_strategy
155
+ )
156
+
157
+
158
+ def concatenate_datasets(
159
+ dsets: List[DatasetType],
160
+ info: Optional[DatasetInfo] = None,
161
+ split: Optional[NamedSplit] = None,
162
+ axis: int = 0,
163
+ ) -> DatasetType:
164
+ """
165
+ Converts a list of [`Dataset`] with the same schema into a single [`Dataset`].
166
+
167
+ Args:
168
+ dsets (`List[datasets.Dataset]`):
169
+ List of Datasets to concatenate.
170
+ info (`DatasetInfo`, *optional*):
171
+ Dataset information, like description, citation, etc.
172
+ split (`NamedSplit`, *optional*):
173
+ Name of the dataset split.
174
+ axis (`{0, 1}`, defaults to `0`):
175
+ Axis to concatenate over, where `0` means over rows (vertically) and `1` means over columns
176
+ (horizontally).
177
+
178
+ <Added version="1.6.0"/>
179
+
180
+ Example:
181
+
182
+ ```py
183
+ >>> ds3 = concatenate_datasets([ds1, ds2])
184
+ ```
185
+ """
186
+
187
+ if not dsets:
188
+ raise ValueError("Unable to concatenate an empty list of datasets.")
189
+ for i, dataset in enumerate(dsets):
190
+ if not isinstance(dataset, (Dataset, IterableDataset)):
191
+ if isinstance(dataset, (DatasetDict, IterableDatasetDict)):
192
+ if not dataset:
193
+ raise ValueError(
194
+ f"Expected a list of Dataset objects or a list of IterableDataset objects, but element at position {i} "
195
+ "is an empty dataset dictionary."
196
+ )
197
+ raise ValueError(
198
+ f"Dataset at position {i} has at least one split: {list(dataset)}\n"
199
+ f"Please pick one to interleave with the other datasets, for example: dataset['{next(iter(dataset))}']"
200
+ )
201
+ raise ValueError(
202
+ f"Expected a list of Dataset objects or a list of IterableDataset objects, but element at position {i} is a {type(dataset).__name__}."
203
+ )
204
+ if i == 0:
205
+ dataset_type, other_type = (
206
+ (Dataset, IterableDataset) if isinstance(dataset, Dataset) else (IterableDataset, Dataset)
207
+ )
208
+ elif not isinstance(dataset, dataset_type):
209
+ raise ValueError(
210
+ f"Unable to interleave a {dataset_type.__name__} (at position 0) with a {other_type.__name__} (at position {i}). Expected a list of Dataset objects or a list of IterableDataset objects."
211
+ )
212
+ if dataset_type is Dataset:
213
+ return _concatenate_map_style_datasets(dsets, info=info, split=split, axis=axis)
214
+ else:
215
+ return _concatenate_iterable_datasets(dsets, info=info, split=split, axis=axis)
venv/lib/python3.10/site-packages/datasets/dataset_dict.py ADDED
The diff for this file is too large to render. See raw diff
 
venv/lib/python3.10/site-packages/datasets/exceptions.py ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # SPDX-License-Identifier: Apache-2.0
2
+ # Copyright 2023 The HuggingFace Authors.
3
+ from typing import Any, Dict, List, Optional, Union
4
+
5
+ from huggingface_hub import HfFileSystem
6
+
7
+ from . import config
8
+ from .table import CastError
9
+ from .utils.track import TrackedIterable, tracked_list, tracked_str
10
+
11
+
12
+ class DatasetsError(Exception):
13
+ """Base class for exceptions in this library."""
14
+
15
+
16
+ class DefunctDatasetError(DatasetsError):
17
+ """The dataset has been defunct."""
18
+
19
+
20
+ class FileNotFoundDatasetsError(DatasetsError, FileNotFoundError):
21
+ """FileNotFoundError raised by this library."""
22
+
23
+
24
+ class DataFilesNotFoundError(FileNotFoundDatasetsError):
25
+ """No (supported) data files found."""
26
+
27
+
28
+ class DatasetNotFoundError(FileNotFoundDatasetsError):
29
+ """Dataset not found.
30
+
31
+ Raised when trying to access:
32
+ - a missing dataset, or
33
+ - a private/gated dataset and the user is not authenticated.
34
+ """
35
+
36
+
37
+ class DatasetBuildError(DatasetsError):
38
+ pass
39
+
40
+
41
+ class ManualDownloadError(DatasetBuildError):
42
+ pass
43
+
44
+
45
+ class FileFormatError(DatasetBuildError):
46
+ pass
47
+
48
+
49
+ class DatasetGenerationError(DatasetBuildError):
50
+ pass
51
+
52
+
53
+ class DatasetGenerationCastError(DatasetGenerationError):
54
+ @classmethod
55
+ def from_cast_error(
56
+ cls,
57
+ cast_error: CastError,
58
+ builder_name: str,
59
+ gen_kwargs: Dict[str, Any],
60
+ token: Optional[Union[bool, str]],
61
+ ) -> "DatasetGenerationCastError":
62
+ explanation_message = (
63
+ f"\n\nAll the data files must have the same columns, but at some point {cast_error.details()}"
64
+ )
65
+ formatted_tracked_gen_kwargs: List[str] = []
66
+ for gen_kwarg in gen_kwargs.values():
67
+ if not isinstance(gen_kwarg, (tracked_str, tracked_list, TrackedIterable)):
68
+ continue
69
+ while isinstance(gen_kwarg, (tracked_list, TrackedIterable)) and gen_kwarg.last_item is not None:
70
+ gen_kwarg = gen_kwarg.last_item
71
+ if isinstance(gen_kwarg, tracked_str):
72
+ gen_kwarg = gen_kwarg.get_origin()
73
+ if isinstance(gen_kwarg, str) and gen_kwarg.startswith("hf://"):
74
+ resolved_path = HfFileSystem(endpoint=config.HF_ENDPOINT, token=token).resolve_path(gen_kwarg)
75
+ gen_kwarg = "hf://" + resolved_path.unresolve()
76
+ if "@" + resolved_path.revision in gen_kwarg:
77
+ gen_kwarg = (
78
+ gen_kwarg.replace("@" + resolved_path.revision, "", 1)
79
+ + f" (at revision {resolved_path.revision})"
80
+ )
81
+ formatted_tracked_gen_kwargs.append(str(gen_kwarg))
82
+ if formatted_tracked_gen_kwargs:
83
+ explanation_message += f"\n\nThis happened while the {builder_name} dataset builder was generating data using\n\n{', '.join(formatted_tracked_gen_kwargs)}"
84
+ help_message = "\n\nPlease either edit the data files to have matching columns, or separate them into different configurations (see docs at https://hf.co/docs/hub/datasets-manual-configuration#multiple-configurations)"
85
+ return cls("An error occurred while generating the dataset" + explanation_message + help_message)
venv/lib/python3.10/site-packages/datasets/inspect.py ADDED
@@ -0,0 +1,582 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Copyright 2020 The HuggingFace Datasets Authors.
2
+ #
3
+ # Licensed under the Apache License, Version 2.0 (the "License");
4
+ # you may not use this file except in compliance with the License.
5
+ # You may obtain a copy of the License at
6
+ #
7
+ # http://www.apache.org/licenses/LICENSE-2.0
8
+ #
9
+ # Unless required by applicable law or agreed to in writing, software
10
+ # distributed under the License is distributed on an "AS IS" BASIS,
11
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12
+ # See the License for the specific language governing permissions and
13
+ # limitations under the License.
14
+
15
+ # Lint as: python3
16
+ """List and inspect datasets."""
17
+
18
+ import inspect
19
+ import os
20
+ import shutil
21
+ import warnings
22
+ from pathlib import Path, PurePath
23
+ from typing import Dict, List, Mapping, Optional, Sequence, Union
24
+
25
+ import huggingface_hub
26
+
27
+ from . import config
28
+ from .download.download_config import DownloadConfig
29
+ from .download.download_manager import DownloadMode
30
+ from .download.streaming_download_manager import StreamingDownloadManager
31
+ from .info import DatasetInfo
32
+ from .load import (
33
+ dataset_module_factory,
34
+ get_dataset_builder_class,
35
+ import_main_class,
36
+ load_dataset_builder,
37
+ metric_module_factory,
38
+ )
39
+ from .utils.deprecation_utils import deprecated
40
+ from .utils.file_utils import relative_to_absolute_path
41
+ from .utils.logging import get_logger
42
+ from .utils.version import Version
43
+
44
+
45
+ logger = get_logger(__name__)
46
+
47
+
48
+ class SplitsNotFoundError(ValueError):
49
+ pass
50
+
51
+
52
+ @deprecated("Use 'huggingface_hub.list_datasets' instead.")
53
+ def list_datasets(with_community_datasets=True, with_details=False):
54
+ """List all the datasets scripts available on the Hugging Face Hub.
55
+
56
+ Args:
57
+ with_community_datasets (`bool`, *optional*, defaults to `True`):
58
+ Include the community provided datasets.
59
+ with_details (`bool`, *optional*, defaults to `False`):
60
+ Return the full details on the datasets instead of only the short name.
61
+
62
+ Example:
63
+
64
+ ```py
65
+ >>> from datasets import list_datasets
66
+ >>> list_datasets()
67
+ ['acronym_identification',
68
+ 'ade_corpus_v2',
69
+ 'adversarial_qa',
70
+ 'aeslc',
71
+ 'afrikaans_ner_corpus',
72
+ 'ag_news',
73
+ ...
74
+ ]
75
+ ```
76
+ """
77
+ datasets = huggingface_hub.list_datasets(full=with_details)
78
+ if not with_community_datasets:
79
+ datasets = [dataset for dataset in datasets if "/" not in dataset.id]
80
+ if not with_details:
81
+ datasets = [dataset.id for dataset in datasets]
82
+ return list(datasets)
83
+
84
+
85
+ @deprecated(
86
+ "Use 'evaluate.list_evaluation_modules' instead, from the new library 🤗 Evaluate: https://huggingface.co/docs/evaluate"
87
+ )
88
+ def list_metrics(with_community_metrics=True, with_details=False):
89
+ """List all the metrics script available on the Hugging Face Hub.
90
+
91
+ <Deprecated version="2.5.0">
92
+
93
+ Use `evaluate.list_evaluation_modules` instead, from the new library 🤗 Evaluate: https://huggingface.co/docs/evaluate
94
+
95
+ </Deprecated>
96
+
97
+ Args:
98
+ with_community_metrics (:obj:`bool`, optional, default ``True``): Include the community provided metrics.
99
+ with_details (:obj:`bool`, optional, default ``False``): Return the full details on the metrics instead of only the short name.
100
+
101
+ Example:
102
+
103
+ ```py
104
+ >>> from datasets import list_metrics
105
+ >>> list_metrics()
106
+ ['accuracy',
107
+ 'bertscore',
108
+ 'bleu',
109
+ 'bleurt',
110
+ 'cer',
111
+ 'chrf',
112
+ ...
113
+ ]
114
+ ```
115
+ """
116
+ metrics = huggingface_hub.list_metrics()
117
+ if not with_community_metrics:
118
+ metrics = [metric for metric in metrics if "/" not in metric.id]
119
+ if not with_details:
120
+ metrics = [metric.id for metric in metrics]
121
+ return metrics
122
+
123
+
124
+ @deprecated("Clone the dataset repository from the Hugging Face Hub instead.")
125
+ def inspect_dataset(path: str, local_path: str, download_config: Optional[DownloadConfig] = None, **download_kwargs):
126
+ """
127
+ Allow inspection/modification of a dataset script by copying on local drive at local_path.
128
+
129
+ Args:
130
+ path (`str`): Path to the dataset processing script with the dataset builder. Can be either:
131
+
132
+ - a local path to processing script or the directory containing the script (if the script has the same name
133
+ as the directory),
134
+ e.g. `'./dataset/squad'` or `'./dataset/squad/squad.py'`.
135
+ - a dataset identifier on the Hugging Face Hub (list all available datasets and ids with [`list_datasets`])
136
+ e.g. `'squad'`, `'glue'` or `'openai/webtext'`.
137
+ local_path (`str`):
138
+ Path to the local folder to copy the dataset script to.
139
+ download_config ([`DownloadConfig`], *optional*):
140
+ Specific download configuration parameters.
141
+ **download_kwargs (additional keyword arguments):
142
+ Optional arguments for [`DownloadConfig`] which will override
143
+ the attributes of `download_config` if supplied.
144
+ """
145
+ if download_config is None:
146
+ download_config = DownloadConfig(**download_kwargs)
147
+ if os.path.isfile(path):
148
+ path = str(Path(path).parent)
149
+ if os.path.isdir(path):
150
+ shutil.copytree(path, local_path, dirs_exist_ok=True)
151
+ else:
152
+ huggingface_hub.HfApi(endpoint=config.HF_ENDPOINT, token=download_config.token).snapshot_download(
153
+ repo_id=path, repo_type="dataset", local_dir=local_path, force_download=download_config.force_download
154
+ )
155
+ print(
156
+ f"The dataset {path} can be inspected at {local_path}. "
157
+ f'You can modify this loading script if it has one and use it with `datasets.load_dataset("{PurePath(local_path).as_posix()}")`.'
158
+ )
159
+
160
+
161
+ @deprecated(
162
+ "Use 'evaluate.inspect_evaluation_module' instead, from the new library 🤗 Evaluate: https://huggingface.co/docs/evaluate"
163
+ )
164
+ def inspect_metric(path: str, local_path: str, download_config: Optional[DownloadConfig] = None, **download_kwargs):
165
+ r"""
166
+ Allow inspection/modification of a metric script by copying it on local drive at local_path.
167
+
168
+ <Deprecated version="2.5.0">
169
+
170
+ Use `evaluate.inspect_evaluation_module` instead, from the new library 🤗 Evaluate instead: https://huggingface.co/docs/evaluate
171
+
172
+ </Deprecated>
173
+
174
+ Args:
175
+ path (``str``): path to the dataset processing script with the dataset builder. Can be either:
176
+
177
+ - a local path to processing script or the directory containing the script (if the script has the same name as the directory),
178
+ e.g. ``'./dataset/squad'`` or ``'./dataset/squad/squad.py'``
179
+ - a dataset identifier on the Hugging Face Hub (list all available datasets and ids with ``datasets.list_datasets()``)
180
+ e.g. ``'squad'``, ``'glue'`` or ``'openai/webtext'``
181
+ local_path (``str``): path to the local folder to copy the datset script to.
182
+ download_config (Optional ``datasets.DownloadConfig``): specific download configuration parameters.
183
+ **download_kwargs (additional keyword arguments): optional attributes for DownloadConfig() which will override the attributes in download_config if supplied.
184
+ """
185
+ metric_module = metric_module_factory(path, download_config=download_config, **download_kwargs)
186
+ metric_cls = import_main_class(metric_module.module_path, dataset=False)
187
+ module_source_path = inspect.getsourcefile(metric_cls)
188
+ module_source_dirpath = os.path.dirname(module_source_path)
189
+ for dirpath, dirnames, filenames in os.walk(module_source_dirpath):
190
+ dst_dirpath = os.path.join(local_path, os.path.relpath(dirpath, module_source_dirpath))
191
+ os.makedirs(dst_dirpath, exist_ok=True)
192
+ # skipping hidden directories; prune the search
193
+ dirnames[:] = [dirname for dirname in dirnames if not dirname.startswith((".", "__"))]
194
+ for filename in filenames:
195
+ shutil.copy2(os.path.join(dirpath, filename), os.path.join(dst_dirpath, filename))
196
+ shutil.copystat(dirpath, dst_dirpath)
197
+ local_path = relative_to_absolute_path(local_path)
198
+ print(
199
+ f"The processing scripts for metric {path} can be inspected at {local_path}. "
200
+ f"The main class is in {module_source_dirpath}. "
201
+ f'You can modify this processing scripts and use it with `datasets.load_metric("{PurePath(local_path).as_posix()}")`.'
202
+ )
203
+
204
+
205
+ def get_dataset_infos(
206
+ path: str,
207
+ data_files: Optional[Union[Dict, List, str]] = None,
208
+ download_config: Optional[DownloadConfig] = None,
209
+ download_mode: Optional[Union[DownloadMode, str]] = None,
210
+ revision: Optional[Union[str, Version]] = None,
211
+ token: Optional[Union[bool, str]] = None,
212
+ use_auth_token="deprecated",
213
+ **config_kwargs,
214
+ ):
215
+ """Get the meta information about a dataset, returned as a dict mapping config name to DatasetInfoDict.
216
+
217
+ Args:
218
+ path (`str`): path to the dataset processing script with the dataset builder. Can be either:
219
+
220
+ - a local path to processing script or the directory containing the script (if the script has the same name as the directory),
221
+ e.g. `'./dataset/squad'` or `'./dataset/squad/squad.py'`
222
+ - a dataset identifier on the Hugging Face Hub (list all available datasets and ids with [`datasets.list_datasets`])
223
+ e.g. `'squad'`, `'glue'` or``'openai/webtext'`
224
+ revision (`Union[str, datasets.Version]`, *optional*):
225
+ If specified, the dataset module will be loaded from the datasets repository at this version.
226
+ By default:
227
+ - it is set to the local version of the lib.
228
+ - it will also try to load it from the main branch if it's not available at the local version of the lib.
229
+ Specifying a version that is different from your local version of the lib might cause compatibility issues.
230
+ download_config ([`DownloadConfig`], *optional*):
231
+ Specific download configuration parameters.
232
+ download_mode ([`DownloadMode`] or `str`, defaults to `REUSE_DATASET_IF_EXISTS`):
233
+ Download/generate mode.
234
+ data_files (`Union[Dict, List, str]`, *optional*):
235
+ Defining the data_files of the dataset configuration.
236
+ token (`str` or `bool`, *optional*):
237
+ Optional string or boolean to use as Bearer token for remote files on the Datasets Hub.
238
+ If `True`, or not specified, will get token from `"~/.huggingface"`.
239
+ use_auth_token (`str` or `bool`, *optional*):
240
+ Optional string or boolean to use as Bearer token for remote files on the Datasets Hub.
241
+ If `True`, or not specified, will get token from `"~/.huggingface"`.
242
+
243
+ <Deprecated version="2.14.0">
244
+
245
+ `use_auth_token` was deprecated in favor of `token` in version 2.14.0 and will be removed in 3.0.0.
246
+
247
+ </Deprecated>
248
+
249
+ **config_kwargs (additional keyword arguments):
250
+ Optional attributes for builder class which will override the attributes if supplied.
251
+
252
+ Example:
253
+
254
+ ```py
255
+ >>> from datasets import get_dataset_infos
256
+ >>> get_dataset_infos('rotten_tomatoes')
257
+ {'default': DatasetInfo(description="Movie Review Dataset.\nThis is a dataset of containing 5,331 positive and 5,331 negative processed\nsentences from Rotten Tomatoes movie reviews...), ...}
258
+ ```
259
+ """
260
+ if use_auth_token != "deprecated":
261
+ warnings.warn(
262
+ "'use_auth_token' was deprecated in favor of 'token' in version 2.14.0 and will be removed in 3.0.0.\n"
263
+ "You can remove this warning by passing 'token=<use_auth_token>' instead.",
264
+ FutureWarning,
265
+ )
266
+ token = use_auth_token
267
+
268
+ config_names = get_dataset_config_names(
269
+ path=path,
270
+ revision=revision,
271
+ download_config=download_config,
272
+ download_mode=download_mode,
273
+ data_files=data_files,
274
+ token=token,
275
+ )
276
+ return {
277
+ config_name: get_dataset_config_info(
278
+ path=path,
279
+ config_name=config_name,
280
+ data_files=data_files,
281
+ download_config=download_config,
282
+ download_mode=download_mode,
283
+ revision=revision,
284
+ token=token,
285
+ **config_kwargs,
286
+ )
287
+ for config_name in config_names
288
+ }
289
+
290
+
291
+ def get_dataset_config_names(
292
+ path: str,
293
+ revision: Optional[Union[str, Version]] = None,
294
+ download_config: Optional[DownloadConfig] = None,
295
+ download_mode: Optional[Union[DownloadMode, str]] = None,
296
+ dynamic_modules_path: Optional[str] = None,
297
+ data_files: Optional[Union[Dict, List, str]] = None,
298
+ **download_kwargs,
299
+ ):
300
+ """Get the list of available config names for a particular dataset.
301
+
302
+ Args:
303
+ path (`str`): path to the dataset processing script with the dataset builder. Can be either:
304
+
305
+ - a local path to processing script or the directory containing the script (if the script has the same name as the directory),
306
+ e.g. `'./dataset/squad'` or `'./dataset/squad/squad.py'`
307
+ - a dataset identifier on the Hugging Face Hub (list all available datasets and ids with [`datasets.list_datasets`])
308
+ e.g. `'squad'`, `'glue'` or `'openai/webtext'`
309
+ revision (`Union[str, datasets.Version]`, *optional*):
310
+ If specified, the dataset module will be loaded from the datasets repository at this version.
311
+ By default:
312
+ - it is set to the local version of the lib.
313
+ - it will also try to load it from the main branch if it's not available at the local version of the lib.
314
+ Specifying a version that is different from your local version of the lib might cause compatibility issues.
315
+ download_config ([`DownloadConfig`], *optional*):
316
+ Specific download configuration parameters.
317
+ download_mode ([`DownloadMode`] or `str`, defaults to `REUSE_DATASET_IF_EXISTS`):
318
+ Download/generate mode.
319
+ dynamic_modules_path (`str`, defaults to `~/.cache/huggingface/modules/datasets_modules`):
320
+ Optional path to the directory in which the dynamic modules are saved. It must have been initialized with `init_dynamic_modules`.
321
+ By default the datasets and metrics are stored inside the `datasets_modules` module.
322
+ data_files (`Union[Dict, List, str]`, *optional*):
323
+ Defining the data_files of the dataset configuration.
324
+ **download_kwargs (additional keyword arguments):
325
+ Optional attributes for [`DownloadConfig`] which will override the attributes in `download_config` if supplied,
326
+ for example `token`.
327
+
328
+ Example:
329
+
330
+ ```py
331
+ >>> from datasets import get_dataset_config_names
332
+ >>> get_dataset_config_names("glue")
333
+ ['cola',
334
+ 'sst2',
335
+ 'mrpc',
336
+ 'qqp',
337
+ 'stsb',
338
+ 'mnli',
339
+ 'mnli_mismatched',
340
+ 'mnli_matched',
341
+ 'qnli',
342
+ 'rte',
343
+ 'wnli',
344
+ 'ax']
345
+ ```
346
+ """
347
+ dataset_module = dataset_module_factory(
348
+ path,
349
+ revision=revision,
350
+ download_config=download_config,
351
+ download_mode=download_mode,
352
+ dynamic_modules_path=dynamic_modules_path,
353
+ data_files=data_files,
354
+ **download_kwargs,
355
+ )
356
+ builder_cls = get_dataset_builder_class(dataset_module, dataset_name=os.path.basename(path))
357
+ return list(builder_cls.builder_configs.keys()) or [
358
+ dataset_module.builder_kwargs.get("config_name", builder_cls.DEFAULT_CONFIG_NAME or "default")
359
+ ]
360
+
361
+
362
+ def get_dataset_default_config_name(
363
+ path: str,
364
+ revision: Optional[Union[str, Version]] = None,
365
+ download_config: Optional[DownloadConfig] = None,
366
+ download_mode: Optional[Union[DownloadMode, str]] = None,
367
+ dynamic_modules_path: Optional[str] = None,
368
+ data_files: Optional[Union[Dict, List, str]] = None,
369
+ **download_kwargs,
370
+ ) -> Optional[str]:
371
+ """Get the default config name for a particular dataset.
372
+ Can return None only if the dataset has multiple configurations and no default configuration.
373
+
374
+ Args:
375
+ path (`str`): path to the dataset processing script with the dataset builder. Can be either:
376
+
377
+ - a local path to processing script or the directory containing the script (if the script has the same name as the directory),
378
+ e.g. `'./dataset/squad'` or `'./dataset/squad/squad.py'`
379
+ - a dataset identifier on the Hugging Face Hub (list all available datasets and ids with [`datasets.list_datasets`])
380
+ e.g. `'squad'`, `'glue'` or `'openai/webtext'`
381
+ revision (`Union[str, datasets.Version]`, *optional*):
382
+ If specified, the dataset module will be loaded from the datasets repository at this version.
383
+ By default:
384
+ - it is set to the local version of the lib.
385
+ - it will also try to load it from the main branch if it's not available at the local version of the lib.
386
+ Specifying a version that is different from your local version of the lib might cause compatibility issues.
387
+ download_config ([`DownloadConfig`], *optional*):
388
+ Specific download configuration parameters.
389
+ download_mode ([`DownloadMode`] or `str`, defaults to `REUSE_DATASET_IF_EXISTS`):
390
+ Download/generate mode.
391
+ dynamic_modules_path (`str`, defaults to `~/.cache/huggingface/modules/datasets_modules`):
392
+ Optional path to the directory in which the dynamic modules are saved. It must have been initialized with `init_dynamic_modules`.
393
+ By default the datasets and metrics are stored inside the `datasets_modules` module.
394
+ data_files (`Union[Dict, List, str]`, *optional*):
395
+ Defining the data_files of the dataset configuration.
396
+ **download_kwargs (additional keyword arguments):
397
+ Optional attributes for [`DownloadConfig`] which will override the attributes in `download_config` if supplied,
398
+ for example `token`.
399
+
400
+ Returns:
401
+ Optional[str]: the default config name if there is one
402
+
403
+ Example:
404
+
405
+ ```py
406
+ >>> from datasets import get_dataset_default_config_name
407
+ >>> get_dataset_default_config_name("openbookqa")
408
+ 'main'
409
+ ```
410
+ """
411
+ dataset_module = dataset_module_factory(
412
+ path,
413
+ revision=revision,
414
+ download_config=download_config,
415
+ download_mode=download_mode,
416
+ dynamic_modules_path=dynamic_modules_path,
417
+ data_files=data_files,
418
+ **download_kwargs,
419
+ )
420
+ builder_cls = get_dataset_builder_class(dataset_module, dataset_name=os.path.basename(path))
421
+ builder_configs = list(builder_cls.builder_configs.keys())
422
+ if builder_configs:
423
+ default_config_name = builder_configs[0] if len(builder_configs) == 1 else None
424
+ else:
425
+ default_config_name = "default"
426
+ return builder_cls.DEFAULT_CONFIG_NAME or default_config_name
427
+
428
+
429
+ def get_dataset_config_info(
430
+ path: str,
431
+ config_name: Optional[str] = None,
432
+ data_files: Optional[Union[str, Sequence[str], Mapping[str, Union[str, Sequence[str]]]]] = None,
433
+ download_config: Optional[DownloadConfig] = None,
434
+ download_mode: Optional[Union[DownloadMode, str]] = None,
435
+ revision: Optional[Union[str, Version]] = None,
436
+ token: Optional[Union[bool, str]] = None,
437
+ use_auth_token="deprecated",
438
+ **config_kwargs,
439
+ ) -> DatasetInfo:
440
+ """Get the meta information (DatasetInfo) about a dataset for a particular config
441
+
442
+ Args:
443
+ path (``str``): path to the dataset processing script with the dataset builder. Can be either:
444
+
445
+ - a local path to processing script or the directory containing the script (if the script has the same name as the directory),
446
+ e.g. ``'./dataset/squad'`` or ``'./dataset/squad/squad.py'``
447
+ - a dataset identifier on the Hugging Face Hub (list all available datasets and ids with ``datasets.list_datasets()``)
448
+ e.g. ``'squad'``, ``'glue'`` or ``'openai/webtext'``
449
+ config_name (:obj:`str`, optional): Defining the name of the dataset configuration.
450
+ data_files (:obj:`str` or :obj:`Sequence` or :obj:`Mapping`, optional): Path(s) to source data file(s).
451
+ download_config (:class:`~download.DownloadConfig`, optional): Specific download configuration parameters.
452
+ download_mode (:class:`DownloadMode` or :obj:`str`, default ``REUSE_DATASET_IF_EXISTS``): Download/generate mode.
453
+ revision (:class:`~utils.Version` or :obj:`str`, optional): Version of the dataset script to load.
454
+ As datasets have their own git repository on the Datasets Hub, the default version "main" corresponds to their "main" branch.
455
+ You can specify a different version than the default "main" by using a commit SHA or a git tag of the dataset repository.
456
+ token (``str`` or :obj:`bool`, optional): Optional string or boolean to use as Bearer token for remote files on the Datasets Hub.
457
+ If True, or not specified, will get token from `"~/.huggingface"`.
458
+ use_auth_token (``str`` or :obj:`bool`, optional): Optional string or boolean to use as Bearer token for remote files on the Datasets Hub.
459
+ If True, or not specified, will get token from `"~/.huggingface"`.
460
+
461
+ <Deprecated version="2.14.0">
462
+
463
+ `use_auth_token` was deprecated in favor of `token` in version 2.14.0 and will be removed in 3.0.0.
464
+
465
+ </Deprecated>
466
+
467
+ **config_kwargs (additional keyword arguments): optional attributes for builder class which will override the attributes if supplied.
468
+
469
+ """
470
+ if use_auth_token != "deprecated":
471
+ warnings.warn(
472
+ "'use_auth_token' was deprecated in favor of 'token' in version 2.14.0 and will be removed in 3.0.0.\n"
473
+ "You can remove this warning by passing 'token=<use_auth_token>' instead.",
474
+ FutureWarning,
475
+ )
476
+ token = use_auth_token
477
+
478
+ builder = load_dataset_builder(
479
+ path,
480
+ name=config_name,
481
+ data_files=data_files,
482
+ download_config=download_config,
483
+ download_mode=download_mode,
484
+ revision=revision,
485
+ token=token,
486
+ **config_kwargs,
487
+ )
488
+ info = builder.info
489
+ if info.splits is None:
490
+ download_config = download_config.copy() if download_config else DownloadConfig()
491
+ if token is not None:
492
+ download_config.token = token
493
+ builder._check_manual_download(
494
+ StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
495
+ )
496
+ try:
497
+ info.splits = {
498
+ split_generator.name: {"name": split_generator.name, "dataset_name": path}
499
+ for split_generator in builder._split_generators(
500
+ StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
501
+ )
502
+ }
503
+ except Exception as err:
504
+ raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
505
+ return info
506
+
507
+
508
+ def get_dataset_split_names(
509
+ path: str,
510
+ config_name: Optional[str] = None,
511
+ data_files: Optional[Union[str, Sequence[str], Mapping[str, Union[str, Sequence[str]]]]] = None,
512
+ download_config: Optional[DownloadConfig] = None,
513
+ download_mode: Optional[Union[DownloadMode, str]] = None,
514
+ revision: Optional[Union[str, Version]] = None,
515
+ token: Optional[Union[bool, str]] = None,
516
+ use_auth_token="deprecated",
517
+ **config_kwargs,
518
+ ):
519
+ """Get the list of available splits for a particular config and dataset.
520
+
521
+ Args:
522
+ path (`str`): path to the dataset processing script with the dataset builder. Can be either:
523
+
524
+ - a local path to processing script or the directory containing the script (if the script has the same name as the directory),
525
+ e.g. `'./dataset/squad'` or `'./dataset/squad/squad.py'`
526
+ - a dataset identifier on the Hugging Face Hub (list all available datasets and ids with [`datasets.list_datasets`])
527
+ e.g. `'squad'`, `'glue'` or `'openai/webtext'`
528
+ config_name (`str`, *optional*):
529
+ Defining the name of the dataset configuration.
530
+ data_files (`str` or `Sequence` or `Mapping`, *optional*):
531
+ Path(s) to source data file(s).
532
+ download_config ([`DownloadConfig`], *optional*):
533
+ Specific download configuration parameters.
534
+ download_mode ([`DownloadMode`] or `str`, defaults to `REUSE_DATASET_IF_EXISTS`):
535
+ Download/generate mode.
536
+ revision ([`Version`] or `str`, *optional*):
537
+ Version of the dataset script to load.
538
+ As datasets have their own git repository on the Datasets Hub, the default version "main" corresponds to their "main" branch.
539
+ You can specify a different version than the default "main" by using a commit SHA or a git tag of the dataset repository.
540
+ token (`str` or `bool`, *optional*):
541
+ Optional string or boolean to use as Bearer token for remote files on the Datasets Hub.
542
+ If `True`, or not specified, will get token from `"~/.huggingface"`.
543
+ use_auth_token (`str` or `bool`, *optional*):
544
+ Optional string or boolean to use as Bearer token for remote files on the Datasets Hub.
545
+ If `True`, or not specified, will get token from `"~/.huggingface"`.
546
+
547
+ <Deprecated version="2.14.0">
548
+
549
+ `use_auth_token` was deprecated in favor of `token` in version 2.14.0 and will be removed in 3.0.0.
550
+
551
+ </Deprecated>
552
+
553
+ **config_kwargs (additional keyword arguments):
554
+ Optional attributes for builder class which will override the attributes if supplied.
555
+
556
+ Example:
557
+
558
+ ```py
559
+ >>> from datasets import get_dataset_split_names
560
+ >>> get_dataset_split_names('rotten_tomatoes')
561
+ ['train', 'validation', 'test']
562
+ ```
563
+ """
564
+ if use_auth_token != "deprecated":
565
+ warnings.warn(
566
+ "'use_auth_token' was deprecated in favor of 'token' in version 2.14.0 and will be removed in 3.0.0.\n"
567
+ "You can remove this warning by passing 'token=<use_auth_token>' instead.",
568
+ FutureWarning,
569
+ )
570
+ token = use_auth_token
571
+
572
+ info = get_dataset_config_info(
573
+ path,
574
+ config_name=config_name,
575
+ data_files=data_files,
576
+ download_config=download_config,
577
+ download_mode=download_mode,
578
+ revision=revision,
579
+ token=token,
580
+ **config_kwargs,
581
+ )
582
+ return list(info.splits.keys())
venv/lib/python3.10/site-packages/datasets/streaming.py ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import importlib
2
+ import inspect
3
+ from functools import wraps
4
+ from typing import TYPE_CHECKING, Optional
5
+
6
+ from .download.download_config import DownloadConfig
7
+ from .utils.file_utils import (
8
+ xbasename,
9
+ xdirname,
10
+ xet_parse,
11
+ xexists,
12
+ xgetsize,
13
+ xglob,
14
+ xgzip_open,
15
+ xisdir,
16
+ xisfile,
17
+ xjoin,
18
+ xlistdir,
19
+ xnumpy_load,
20
+ xopen,
21
+ xpandas_read_csv,
22
+ xpandas_read_excel,
23
+ xPath,
24
+ xpyarrow_parquet_read_table,
25
+ xrelpath,
26
+ xsio_loadmat,
27
+ xsplit,
28
+ xsplitext,
29
+ xwalk,
30
+ xxml_dom_minidom_parse,
31
+ )
32
+ from .utils.logging import get_logger
33
+ from .utils.patching import patch_submodule
34
+ from .utils.py_utils import get_imports, lock_importable_file
35
+
36
+
37
+ logger = get_logger(__name__)
38
+
39
+
40
+ if TYPE_CHECKING:
41
+ from .builder import DatasetBuilder
42
+
43
+
44
+ def extend_module_for_streaming(module_path, download_config: Optional[DownloadConfig] = None):
45
+ """Extend the module to support streaming.
46
+
47
+ We patch some functions in the module to use `fsspec` to support data streaming:
48
+ - We use `fsspec.open` to open and read remote files. We patch the module function:
49
+ - `open`
50
+ - We use the "::" hop separator to join paths and navigate remote compressed/archive files. We patch the module
51
+ functions:
52
+ - `os.path.join`
53
+ - `pathlib.Path.joinpath` and `pathlib.Path.__truediv__` (called when using the "/" operator)
54
+
55
+ The patched functions are replaced with custom functions defined to work with the
56
+ :class:`~download.streaming_download_manager.StreamingDownloadManager`.
57
+
58
+ Args:
59
+ module_path: Path to the module to be extended.
60
+ download_config : mainly use use_auth_token or storage_options to support different platforms and auth types.
61
+ """
62
+
63
+ module = importlib.import_module(module_path)
64
+
65
+ # TODO(QL): always update the module to add subsequent new authentication without removing old ones
66
+ if hasattr(module, "_patched_for_streaming") and module._patched_for_streaming:
67
+ if isinstance(module._patched_for_streaming, DownloadConfig):
68
+ module._patched_for_streaming.token = download_config.token
69
+ module._patched_for_streaming.storage_options = download_config.storage_options
70
+ return
71
+
72
+ def wrap_auth(function):
73
+ @wraps(function)
74
+ def wrapper(*args, **kwargs):
75
+ return function(*args, download_config=download_config, **kwargs)
76
+
77
+ wrapper._decorator_name_ = "wrap_auth"
78
+ return wrapper
79
+
80
+ # open files in a streaming fashion
81
+ patch_submodule(module, "open", wrap_auth(xopen)).start()
82
+ patch_submodule(module, "os.listdir", wrap_auth(xlistdir)).start()
83
+ patch_submodule(module, "os.walk", wrap_auth(xwalk)).start()
84
+ patch_submodule(module, "glob.glob", wrap_auth(xglob)).start()
85
+ # allow to navigate in remote zip files
86
+ patch_submodule(module, "os.path.join", xjoin).start()
87
+ patch_submodule(module, "os.path.dirname", xdirname).start()
88
+ patch_submodule(module, "os.path.basename", xbasename).start()
89
+ patch_submodule(module, "os.path.relpath", xrelpath).start()
90
+ patch_submodule(module, "os.path.split", xsplit).start()
91
+ patch_submodule(module, "os.path.splitext", xsplitext).start()
92
+ # allow checks on paths
93
+ patch_submodule(module, "os.path.exists", wrap_auth(xexists)).start()
94
+ patch_submodule(module, "os.path.isdir", wrap_auth(xisdir)).start()
95
+ patch_submodule(module, "os.path.isfile", wrap_auth(xisfile)).start()
96
+ patch_submodule(module, "os.path.getsize", wrap_auth(xgetsize)).start()
97
+ patch_submodule(module, "pathlib.Path", xPath).start()
98
+ # file readers
99
+ patch_submodule(module, "gzip.open", wrap_auth(xgzip_open)).start()
100
+ patch_submodule(module, "numpy.load", wrap_auth(xnumpy_load)).start()
101
+ patch_submodule(module, "pandas.read_csv", wrap_auth(xpandas_read_csv), attrs=["__version__"]).start()
102
+ patch_submodule(module, "pandas.read_excel", wrap_auth(xpandas_read_excel), attrs=["__version__"]).start()
103
+ patch_submodule(module, "scipy.io.loadmat", wrap_auth(xsio_loadmat), attrs=["__version__"]).start()
104
+ patch_submodule(module, "xml.etree.ElementTree.parse", wrap_auth(xet_parse)).start()
105
+ patch_submodule(module, "xml.dom.minidom.parse", wrap_auth(xxml_dom_minidom_parse)).start()
106
+ # pyarrow: do not patch pyarrow attribute in packaged modules
107
+ if not module.__name__.startswith("datasets.packaged_modules."):
108
+ patch_submodule(module, "pyarrow.parquet.read_table", wrap_auth(xpyarrow_parquet_read_table)).start()
109
+ module._patched_for_streaming = download_config
110
+
111
+
112
+ def extend_dataset_builder_for_streaming(builder: "DatasetBuilder"):
113
+ """Extend the dataset builder module and the modules imported by it to support streaming.
114
+
115
+ Args:
116
+ builder (:class:`DatasetBuilder`): Dataset builder instance.
117
+ """
118
+ # this extends the open and os.path.join functions for data streaming
119
+ download_config = DownloadConfig(storage_options=builder.storage_options, token=builder.token)
120
+ extend_module_for_streaming(builder.__module__, download_config=download_config)
121
+ # if needed, we also have to extend additional internal imports (like wmt14 -> wmt_utils)
122
+ if not builder.__module__.startswith("datasets."): # check that it's not a packaged builder like csv
123
+ importable_file = inspect.getfile(builder.__class__)
124
+ with lock_importable_file(importable_file):
125
+ for imports in get_imports(importable_file):
126
+ if imports[0] == "internal":
127
+ internal_import_name = imports[1]
128
+ internal_module_name = ".".join(builder.__module__.split(".")[:-1] + [internal_import_name])
129
+ extend_module_for_streaming(internal_module_name, download_config=download_config)
130
+
131
+ # builders can inherit from other builders that might use streaming functionality
132
+ # (for example, ImageFolder and AudioFolder inherit from FolderBuilder which implements examples generation)
133
+ # but these parents builders are not patched automatically as they are not instantiated, so we patch them here
134
+ from .builder import DatasetBuilder
135
+
136
+ parent_builder_modules = [
137
+ cls.__module__
138
+ for cls in type(builder).__mro__[1:] # make sure it's not the same module we've already patched
139
+ if issubclass(cls, DatasetBuilder) and cls.__module__ != DatasetBuilder.__module__
140
+ ] # check it's not a standard builder from datasets.builder
141
+ for module in parent_builder_modules:
142
+ extend_module_for_streaming(module, download_config=download_config)
venv/lib/python3.10/site-packages/datasets/table.py ADDED
@@ -0,0 +1,2415 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import copy
2
+ import os
3
+ from functools import partial
4
+ from itertools import groupby
5
+ from typing import TYPE_CHECKING, Callable, Iterator, List, Optional, Tuple, TypeVar, Union
6
+
7
+ import numpy as np
8
+ import pyarrow as pa
9
+ import pyarrow.compute as pc
10
+ import pyarrow.types
11
+
12
+ from . import config
13
+ from .utils.logging import get_logger
14
+
15
+
16
+ if TYPE_CHECKING:
17
+ from .features.features import Features, FeatureType
18
+
19
+
20
+ logger = get_logger(__name__)
21
+
22
+
23
+ def inject_arrow_table_documentation(arrow_table_method):
24
+ def wrapper(fn):
25
+ fn.__doc__ = arrow_table_method.__doc__ + (fn.__doc__ if fn.__doc__ is not None else "")
26
+ fn.__doc__ = fn.__doc__.replace("pyarrow.Table", "Table")
27
+ if hasattr(arrow_table_method, "__annotations__"):
28
+ fn.__annotations__ = arrow_table_method.__annotations__
29
+ return fn
30
+
31
+ return wrapper
32
+
33
+
34
+ def _in_memory_arrow_table_from_file(filename: str) -> pa.Table:
35
+ in_memory_stream = pa.input_stream(filename)
36
+ opened_stream = pa.ipc.open_stream(in_memory_stream)
37
+ pa_table = opened_stream.read_all()
38
+ return pa_table
39
+
40
+
41
+ def _in_memory_arrow_table_from_buffer(buffer: pa.Buffer) -> pa.Table:
42
+ stream = pa.BufferReader(buffer)
43
+ opened_stream = pa.ipc.open_stream(stream)
44
+ table = opened_stream.read_all()
45
+ return table
46
+
47
+
48
+ def _memory_mapped_record_batch_reader_from_file(filename: str) -> pa.RecordBatchStreamReader:
49
+ memory_mapped_stream = pa.memory_map(filename)
50
+ return pa.ipc.open_stream(memory_mapped_stream)
51
+
52
+
53
+ def read_schema_from_file(filename: str) -> pa.Schema:
54
+ """
55
+ Infer arrow table schema from file without loading whole file into memory.
56
+ Usefull especially while having very big files.
57
+ """
58
+ with pa.memory_map(filename) as memory_mapped_stream:
59
+ schema = pa.ipc.open_stream(memory_mapped_stream).schema
60
+ return schema
61
+
62
+
63
+ def _memory_mapped_arrow_table_from_file(filename: str) -> pa.Table:
64
+ opened_stream = _memory_mapped_record_batch_reader_from_file(filename)
65
+ pa_table = opened_stream.read_all()
66
+ return pa_table
67
+
68
+
69
+ def _deepcopy(x, memo: dict):
70
+ """deepcopy a regular class instance"""
71
+ cls = x.__class__
72
+ result = cls.__new__(cls)
73
+ memo[id(x)] = result
74
+ for k, v in x.__dict__.items():
75
+ setattr(result, k, copy.deepcopy(v, memo))
76
+ return result
77
+
78
+
79
+ def _interpolation_search(arr: List[int], x: int) -> int:
80
+ """
81
+ Return the position i of a sorted array so that arr[i] <= x < arr[i+1]
82
+
83
+ Args:
84
+ arr (`List[int]`): non-empty sorted list of integers
85
+ x (`int`): query
86
+
87
+ Returns:
88
+ `int`: the position i so that arr[i] <= x < arr[i+1]
89
+
90
+ Raises:
91
+ `IndexError`: if the array is empty or if the query is outside the array values
92
+ """
93
+ i, j = 0, len(arr) - 1
94
+ while i < j and arr[i] <= x < arr[j]:
95
+ k = i + ((j - i) * (x - arr[i]) // (arr[j] - arr[i]))
96
+ if arr[k] <= x < arr[k + 1]:
97
+ return k
98
+ elif arr[k] < x:
99
+ i, j = k + 1, j
100
+ else:
101
+ i, j = i, k
102
+ raise IndexError(f"Invalid query '{x}' for size {arr[-1] if len(arr) else 'none'}.")
103
+
104
+
105
+ class IndexedTableMixin:
106
+ def __init__(self, table: pa.Table):
107
+ self._schema: pa.Schema = table.schema
108
+ self._batches: List[pa.RecordBatch] = [
109
+ recordbatch for recordbatch in table.to_batches() if len(recordbatch) > 0
110
+ ]
111
+ self._offsets: np.ndarray = np.cumsum([0] + [len(b) for b in self._batches], dtype=np.int64)
112
+
113
+ def fast_gather(self, indices: Union[List[int], np.ndarray]) -> pa.Table:
114
+ """
115
+ Create a pa.Table by gathering the records at the records at the specified indices. Should be faster
116
+ than pa.concat_tables(table.fast_slice(int(i) % table.num_rows, 1) for i in indices) since NumPy can compute
117
+ the binary searches in parallel, highly optimized C
118
+ """
119
+ if not len(indices):
120
+ raise ValueError("Indices must be non-empty")
121
+ batch_indices = np.searchsorted(self._offsets, indices, side="right") - 1
122
+ return pa.Table.from_batches(
123
+ [
124
+ self._batches[batch_idx].slice(i - self._offsets[batch_idx], 1)
125
+ for batch_idx, i in zip(batch_indices, indices)
126
+ ],
127
+ schema=self._schema,
128
+ )
129
+
130
+ def fast_slice(self, offset=0, length=None) -> pa.Table:
131
+ """
132
+ Slice the Table using interpolation search.
133
+ The behavior is the same as `pyarrow.Table.slice` but it's significantly faster.
134
+
135
+ Interpolation search is used to find the start and end indexes of the batches we want to keep.
136
+ The batches to keep are then concatenated to form the sliced Table.
137
+ """
138
+ if offset < 0:
139
+ raise IndexError("Offset must be non-negative")
140
+ elif offset >= self._offsets[-1] or (length is not None and length <= 0):
141
+ return pa.Table.from_batches([], schema=self._schema)
142
+ i = _interpolation_search(self._offsets, offset)
143
+ if length is None or length + offset >= self._offsets[-1]:
144
+ batches = self._batches[i:]
145
+ batches[0] = batches[0].slice(offset - self._offsets[i])
146
+ else:
147
+ j = _interpolation_search(self._offsets, offset + length - 1)
148
+ batches = self._batches[i : j + 1]
149
+ batches[-1] = batches[-1].slice(0, offset + length - self._offsets[j])
150
+ batches[0] = batches[0].slice(offset - self._offsets[i])
151
+ return pa.Table.from_batches(batches, schema=self._schema)
152
+
153
+
154
+ class Table(IndexedTableMixin):
155
+ """
156
+ Wraps a pyarrow Table by using composition.
157
+ This is the base class for `InMemoryTable`, `MemoryMappedTable` and `ConcatenationTable`.
158
+
159
+ It implements all the basic attributes/methods of the pyarrow Table class except
160
+ the Table transforms: `slice, filter, flatten, combine_chunks, cast, add_column,
161
+ append_column, remove_column, set_column, rename_columns` and `drop`.
162
+
163
+ The implementation of these methods differs for the subclasses.
164
+ """
165
+
166
+ def __init__(self, table: pa.Table):
167
+ super().__init__(table)
168
+ self.table = table
169
+
170
+ def __deepcopy__(self, memo: dict):
171
+ # arrow tables are immutable, so there's no need to copy self.table
172
+ # moreover calling deepcopy on a pyarrow table seems to make pa.total_allocated_bytes() decrease for some reason
173
+ # by adding it to the memo, self.table won't be copied
174
+ memo[id(self.table)] = self.table
175
+ # same for the recordbatches used by the index
176
+ memo[id(self._batches)] = list(self._batches)
177
+ return _deepcopy(self, memo)
178
+
179
+ def validate(self, *args, **kwargs):
180
+ """
181
+ Perform validation checks. An exception is raised if validation fails.
182
+
183
+ By default only cheap validation checks are run. Pass `full=True`
184
+ for thorough validation checks (potentially `O(n)`).
185
+
186
+ Args:
187
+ full (`bool`, defaults to `False`):
188
+ If `True`, run expensive checks, otherwise cheap checks only.
189
+
190
+ Raises:
191
+ `pa.lib.ArrowInvalid`: if validation fails
192
+ """
193
+ return self.table.validate(*args, **kwargs)
194
+
195
+ def equals(self, *args, **kwargs):
196
+ """
197
+ Check if contents of two tables are equal.
198
+
199
+ Args:
200
+ other ([`~datasets.table.Table`]):
201
+ Table to compare against.
202
+ check_metadata `bool`, defaults to `False`):
203
+ Whether schema metadata equality should be checked as well.
204
+
205
+ Returns:
206
+ `bool`
207
+ """
208
+ args = tuple(arg.table if isinstance(arg, Table) else arg for arg in args)
209
+ kwargs = {k: v.table if isinstance(v, Table) else v for k, v in kwargs}
210
+ return self.table.equals(*args, **kwargs)
211
+
212
+ def to_batches(self, *args, **kwargs):
213
+ """
214
+ Convert Table to list of (contiguous) `RecordBatch` objects.
215
+
216
+ Args:
217
+ max_chunksize (`int`, defaults to `None`):
218
+ Maximum size for `RecordBatch` chunks. Individual chunks may be
219
+ smaller depending on the chunk layout of individual columns.
220
+
221
+ Returns:
222
+ `List[pyarrow.RecordBatch]`
223
+ """
224
+ return self.table.to_batches(*args, **kwargs)
225
+
226
+ def to_pydict(self, *args, **kwargs):
227
+ """
228
+ Convert the Table to a `dict` or `OrderedDict`.
229
+
230
+ Returns:
231
+ `dict`
232
+ """
233
+ return self.table.to_pydict(*args, **kwargs)
234
+
235
+ def to_pylist(self, *args, **kwargs):
236
+ """
237
+ Convert the Table to a list
238
+
239
+ Returns:
240
+ `list`
241
+ """
242
+ return self.table.to_pylist(*args, **kwargs)
243
+
244
+ def to_pandas(self, *args, **kwargs):
245
+ """
246
+ Convert to a pandas-compatible NumPy array or DataFrame, as appropriate.
247
+
248
+ Args:
249
+ memory_pool (`MemoryPool`, defaults to `None`):
250
+ Arrow MemoryPool to use for allocations. Uses the default memory
251
+ pool is not passed.
252
+ strings_to_categorical (`bool`, defaults to `False`):
253
+ Encode string (UTF8) and binary types to `pandas.Categorical`.
254
+ categories (`list`, defaults to `empty`):
255
+ List of fields that should be returned as `pandas.Categorical`. Only
256
+ applies to table-like data structures.
257
+ zero_copy_only (`bool`, defaults to `False`):
258
+ Raise an `ArrowException` if this function call would require copying
259
+ the underlying data.
260
+ integer_object_nulls (`bool`, defaults to `False`):
261
+ Cast integers with nulls to objects.
262
+ date_as_object (`bool`, defaults to `True`):
263
+ Cast dates to objects. If `False`, convert to `datetime64[ns]` dtype.
264
+ timestamp_as_object (`bool`, defaults to `False`):
265
+ Cast non-nanosecond timestamps (`np.datetime64`) to objects. This is
266
+ useful if you have timestamps that don't fit in the normal date
267
+ range of nanosecond timestamps (1678 CE-2262 CE).
268
+ If `False`, all timestamps are converted to `datetime64[ns]` dtype.
269
+ use_threads (`bool`, defaults to `True`):
270
+ Whether to parallelize the conversion using multiple threads.
271
+ deduplicate_objects (`bool`, defaults to `False`):
272
+ Do not create multiple copies Python objects when created, to save
273
+ on memory use. Conversion will be slower.
274
+ ignore_metadata (`bool`, defaults to `False`):
275
+ If `True`, do not use the 'pandas' metadata to reconstruct the
276
+ DataFrame index, if present.
277
+ safe (`bool`, defaults to `True`):
278
+ For certain data types, a cast is needed in order to store the
279
+ data in a pandas DataFrame or Series (e.g. timestamps are always
280
+ stored as nanoseconds in pandas). This option controls whether it
281
+ is a safe cast or not.
282
+ split_blocks (`bool`, defaults to `False`):
283
+ If `True`, generate one internal "block" for each column when
284
+ creating a pandas.DataFrame from a `RecordBatch` or `Table`. While this
285
+ can temporarily reduce memory note that various pandas operations
286
+ can trigger "consolidation" which may balloon memory use.
287
+ self_destruct (`bool`, defaults to `False`):
288
+ EXPERIMENTAL: If `True`, attempt to deallocate the originating Arrow
289
+ memory while converting the Arrow object to pandas. If you use the
290
+ object after calling `to_pandas` with this option it will crash your
291
+ program.
292
+ types_mapper (`function`, defaults to `None`):
293
+ A function mapping a pyarrow DataType to a pandas `ExtensionDtype`.
294
+ This can be used to override the default pandas type for conversion
295
+ of built-in pyarrow types or in absence of `pandas_metadata` in the
296
+ Table schema. The function receives a pyarrow DataType and is
297
+ expected to return a pandas `ExtensionDtype` or `None` if the
298
+ default conversion should be used for that type. If you have
299
+ a dictionary mapping, you can pass `dict.get` as function.
300
+
301
+ Returns:
302
+ `pandas.Series` or `pandas.DataFrame`: `pandas.Series` or `pandas.DataFrame` depending on type of object
303
+ """
304
+ return self.table.to_pandas(*args, **kwargs)
305
+
306
+ def to_string(self, *args, **kwargs):
307
+ return self.table.to_string(*args, **kwargs)
308
+
309
+ def to_reader(self, max_chunksize: Optional[int] = None):
310
+ """
311
+ Convert the Table to a RecordBatchReader.
312
+
313
+ Note that this method is zero-copy, it merely exposes the same data under a different API.
314
+
315
+ Args:
316
+ max_chunksize (`int`, defaults to `None`)
317
+ Maximum size for RecordBatch chunks. Individual chunks may be smaller depending
318
+ on the chunk layout of individual columns.
319
+
320
+ Returns:
321
+ `pyarrow.RecordBatchReader`
322
+ """
323
+ return self.table.to_reader(max_chunksize=max_chunksize)
324
+
325
+ def field(self, *args, **kwargs):
326
+ """
327
+ Select a schema field by its column name or numeric index.
328
+
329
+ Args:
330
+ i (`Union[int, str]`):
331
+ The index or name of the field to retrieve.
332
+
333
+ Returns:
334
+ `pyarrow.Field`
335
+ """
336
+ return self.table.field(*args, **kwargs)
337
+
338
+ def column(self, *args, **kwargs):
339
+ """
340
+ Select a column by its column name, or numeric index.
341
+
342
+ Args:
343
+ i (`Union[int, str]`):
344
+ The index or name of the column to retrieve.
345
+
346
+ Returns:
347
+ `pyarrow.ChunkedArray`
348
+ """
349
+ return self.table.column(*args, **kwargs)
350
+
351
+ def itercolumns(self, *args, **kwargs):
352
+ """
353
+ Iterator over all columns in their numerical order.
354
+
355
+ Yields:
356
+ `pyarrow.ChunkedArray`
357
+ """
358
+ return self.table.itercolumns(*args, **kwargs)
359
+
360
+ @property
361
+ def schema(self):
362
+ """
363
+ Schema of the table and its columns.
364
+
365
+ Returns:
366
+ `pyarrow.Schema`
367
+ """
368
+ return self.table.schema
369
+
370
+ @property
371
+ def columns(self):
372
+ """
373
+ List of all columns in numerical order.
374
+
375
+ Returns:
376
+ `List[pa.ChunkedArray]`
377
+ """
378
+ return self.table.columns
379
+
380
+ @property
381
+ def num_columns(self):
382
+ """
383
+ Number of columns in this table.
384
+
385
+ Returns:
386
+ int
387
+ """
388
+ return self.table.num_columns
389
+
390
+ @property
391
+ def num_rows(self):
392
+ """
393
+ Number of rows in this table.
394
+
395
+ Due to the definition of a table, all columns have the same number of
396
+ rows.
397
+
398
+ Returns:
399
+ int
400
+ """
401
+ return self.table.num_rows
402
+
403
+ @property
404
+ def shape(self):
405
+ """
406
+ Dimensions of the table: (#rows, #columns).
407
+
408
+ Returns:
409
+ `(int, int)`: Number of rows and number of columns.
410
+ """
411
+ return self.table.shape
412
+
413
+ @property
414
+ def nbytes(self):
415
+ """
416
+ Total number of bytes consumed by the elements of the table.
417
+ """
418
+ return self.table.nbytes
419
+
420
+ @property
421
+ def column_names(self):
422
+ """
423
+ Names of the table's columns.
424
+ """
425
+ return self.table.column_names
426
+
427
+ def __eq__(self, other):
428
+ return self.equals(other)
429
+
430
+ def __getitem__(self, i):
431
+ return self.table[i]
432
+
433
+ def __len__(self):
434
+ return len(self.table)
435
+
436
+ def __repr__(self):
437
+ return self.table.__repr__().replace("pyarrow.Table", self.__class__.__name__)
438
+
439
+ def __str__(self):
440
+ return self.table.__str__().replace("pyarrow.Table", self.__class__.__name__)
441
+
442
+ def slice(self, *args, **kwargs):
443
+ """
444
+ Compute zero-copy slice of this Table.
445
+
446
+ Args:
447
+ offset (`int`, defaults to `0`):
448
+ Offset from start of table to slice.
449
+ length (`int`, defaults to `None`):
450
+ Length of slice (default is until end of table starting from
451
+ offset).
452
+
453
+ Returns:
454
+ `datasets.table.Table`
455
+ """
456
+ raise NotImplementedError()
457
+
458
+ def filter(self, *args, **kwargs):
459
+ """
460
+ Select records from a Table. See `pyarrow.compute.filter` for full usage.
461
+ """
462
+ raise NotImplementedError()
463
+
464
+ def flatten(self, *args, **kwargs):
465
+ """
466
+ Flatten this Table. Each column with a struct type is flattened
467
+ into one column per struct field. Other columns are left unchanged.
468
+
469
+ Args:
470
+ memory_pool (`MemoryPool`, defaults to `None`):
471
+ For memory allocations, if required, otherwise use default pool.
472
+
473
+ Returns:
474
+ `datasets.table.Table`
475
+ """
476
+ raise NotImplementedError()
477
+
478
+ def combine_chunks(self, *args, **kwargs):
479
+ """
480
+ Make a new table by combining the chunks this table has.
481
+
482
+ All the underlying chunks in the `ChunkedArray` of each column are
483
+ concatenated into zero or one chunk.
484
+
485
+ Args:
486
+ memory_pool (`MemoryPool`, defaults to `None`):
487
+ For memory allocations, if required, otherwise use default pool.
488
+
489
+ Returns:
490
+ `datasets.table.Table`
491
+ """
492
+ raise NotImplementedError()
493
+
494
+ def cast(self, *args, **kwargs):
495
+ """
496
+ Cast table values to another schema.
497
+
498
+ Args:
499
+ target_schema (`Schema`):
500
+ Schema to cast to, the names and order of fields must match.
501
+ safe (`bool`, defaults to `True`):
502
+ Check for overflows or other unsafe conversions.
503
+
504
+ Returns:
505
+ `datasets.table.Table`
506
+ """
507
+ raise NotImplementedError()
508
+
509
+ def replace_schema_metadata(self, *args, **kwargs):
510
+ """
511
+ EXPERIMENTAL: Create shallow copy of table by replacing schema
512
+ key-value metadata with the indicated new metadata (which may be None,
513
+ which deletes any existing metadata
514
+
515
+ Args:
516
+ metadata (`dict`, defaults to `None`):
517
+
518
+ Returns:
519
+ `datasets.table.Table`: shallow_copy
520
+ """
521
+ raise NotImplementedError()
522
+
523
+ def add_column(self, *args, **kwargs):
524
+ """
525
+ Add column to Table at position.
526
+
527
+ A new table is returned with the column added, the original table
528
+ object is left unchanged.
529
+
530
+ Args:
531
+ i (`int`):
532
+ Index to place the column at.
533
+ field_ (`Union[str, pyarrow.Field]`):
534
+ If a string is passed then the type is deduced from the column
535
+ data.
536
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
537
+ Column data.
538
+
539
+ Returns:
540
+ `datasets.table.Table`: New table with the passed column added.
541
+ """
542
+ raise NotImplementedError()
543
+
544
+ def append_column(self, *args, **kwargs):
545
+ """
546
+ Append column at end of columns.
547
+
548
+ Args:
549
+ field_ (`Union[str, pyarrow.Field]`):
550
+ If a string is passed then the type is deduced from the column
551
+ data.
552
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
553
+ Column data.
554
+
555
+ Returns:
556
+ `datasets.table.Table`: New table with the passed column added.
557
+ """
558
+ raise NotImplementedError()
559
+
560
+ def remove_column(self, *args, **kwargs):
561
+ """
562
+ Create new Table with the indicated column removed.
563
+
564
+ Args:
565
+ i (`int`):
566
+ Index of column to remove.
567
+
568
+ Returns:
569
+ `datasets.table.Table`: New table without the column.
570
+ """
571
+ raise NotImplementedError()
572
+
573
+ def set_column(self, *args, **kwargs):
574
+ """
575
+ Replace column in Table at position.
576
+
577
+ Args:
578
+ i (`int`):
579
+ Index to place the column at.
580
+ field_ (`Union[str, pyarrow.Field]`):
581
+ If a string is passed then the type is deduced from the column
582
+ data.
583
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
584
+ Column data.
585
+
586
+ Returns:
587
+ `datasets.table.Table`: New table with the passed column set.
588
+ """
589
+ raise NotImplementedError()
590
+
591
+ def rename_columns(self, *args, **kwargs):
592
+ """
593
+ Create new table with columns renamed to provided names.
594
+ """
595
+ raise NotImplementedError()
596
+
597
+ def drop(self, *args, **kwargs):
598
+ """
599
+ Drop one or more columns and return a new table.
600
+
601
+ Args:
602
+ columns (`List[str]`):
603
+ List of field names referencing existing columns.
604
+
605
+ Raises:
606
+ `KeyError` : if any of the passed columns name are not existing.
607
+
608
+ Returns:
609
+ `datasets.table.Table`: New table without the columns.
610
+ """
611
+ raise NotImplementedError()
612
+
613
+ def select(self, *args, **kwargs):
614
+ """
615
+ Select columns of the table.
616
+
617
+ Returns a new table with the specified columns, and metadata preserved.
618
+
619
+ Args:
620
+ columns (:obj:`Union[List[str], List[int]]`):
621
+ The column names or integer indices to select.
622
+
623
+ Returns:
624
+ `datasets.table.Table`: table with only a subset of the columns
625
+ """
626
+ raise NotImplementedError()
627
+
628
+
629
+ class TableBlock(Table):
630
+ """
631
+ `TableBlock` is the allowed class inside a `ConcanetationTable`.
632
+ Only `MemoryMappedTable` and `InMemoryTable` are `TableBlock`.
633
+ This is because we don't want a `ConcanetationTable` made out of other `ConcanetationTables`.
634
+ """
635
+
636
+ pass
637
+
638
+
639
+ class InMemoryTable(TableBlock):
640
+ """
641
+ The table is said in-memory when it is loaded into the user's RAM.
642
+
643
+ Pickling it does copy all the data using memory.
644
+ Its implementation is simple and uses the underlying pyarrow Table methods directly.
645
+
646
+ This is different from the `MemoryMapped` table, for which pickling doesn't copy all the
647
+ data in memory. For a `MemoryMapped`, unpickling instead reloads the table from the disk.
648
+
649
+ `InMemoryTable` must be used when data fit in memory, while `MemoryMapped` are reserved for
650
+ data bigger than memory or when you want the memory footprint of your application to
651
+ stay low.
652
+ """
653
+
654
+ @classmethod
655
+ def from_file(cls, filename: str):
656
+ table = _in_memory_arrow_table_from_file(filename)
657
+ return cls(table)
658
+
659
+ @classmethod
660
+ def from_buffer(cls, buffer: pa.Buffer):
661
+ table = _in_memory_arrow_table_from_buffer(buffer)
662
+ return cls(table)
663
+
664
+ @classmethod
665
+ def from_pandas(cls, *args, **kwargs):
666
+ """
667
+ Convert pandas.DataFrame to an Arrow Table.
668
+
669
+ The column types in the resulting Arrow Table are inferred from the
670
+ dtypes of the pandas.Series in the DataFrame. In the case of non-object
671
+ Series, the NumPy dtype is translated to its Arrow equivalent. In the
672
+ case of `object`, we need to guess the datatype by looking at the
673
+ Python objects in this Series.
674
+
675
+ Be aware that Series of the `object` dtype don't carry enough
676
+ information to always lead to a meaningful Arrow type. In the case that
677
+ we cannot infer a type, e.g. because the DataFrame is of length 0 or
678
+ the Series only contains `None/nan` objects, the type is set to
679
+ null. This behavior can be avoided by constructing an explicit schema
680
+ and passing it to this function.
681
+
682
+ Args:
683
+ df (`pandas.DataFrame`):
684
+ schema (`pyarrow.Schema`, *optional*):
685
+ The expected schema of the Arrow Table. This can be used to
686
+ indicate the type of columns if we cannot infer it automatically.
687
+ If passed, the output will have exactly this schema. Columns
688
+ specified in the schema that are not found in the DataFrame columns
689
+ or its index will raise an error. Additional columns or index
690
+ levels in the DataFrame which are not specified in the schema will
691
+ be ignored.
692
+ preserve_index (`bool`, *optional*):
693
+ Whether to store the index as an additional column in the resulting
694
+ `Table`. The default of None will store the index as a column,
695
+ except for RangeIndex which is stored as metadata only. Use
696
+ `preserve_index=True` to force it to be stored as a column.
697
+ nthreads (`int`, defaults to `None` (may use up to system CPU count threads))
698
+ If greater than 1, convert columns to Arrow in parallel using
699
+ indicated number of threads.
700
+ columns (`List[str]`, *optional*):
701
+ List of column to be converted. If `None`, use all columns.
702
+ safe (`bool`, defaults to `True`):
703
+ Check for overflows or other unsafe conversions,
704
+
705
+ Returns:
706
+ `datasets.table.Table`:
707
+
708
+ Examples:
709
+ ```python
710
+ >>> import pandas as pd
711
+ >>> import pyarrow as pa
712
+ >>> df = pd.DataFrame({
713
+ ... 'int': [1, 2],
714
+ ... 'str': ['a', 'b']
715
+ ... })
716
+ >>> pa.Table.from_pandas(df)
717
+ <pyarrow.lib.Table object at 0x7f05d1fb1b40>
718
+ ```
719
+ """
720
+ return cls(pa.Table.from_pandas(*args, **kwargs))
721
+
722
+ @classmethod
723
+ def from_arrays(cls, *args, **kwargs):
724
+ """
725
+ Construct a Table from Arrow arrays.
726
+
727
+ Args:
728
+ arrays (`List[Union[pyarrow.Array, pyarrow.ChunkedArray]]`):
729
+ Equal-length arrays that should form the table.
730
+ names (`List[str]`, *optional*):
731
+ Names for the table columns. If not passed, schema must be passed.
732
+ schema (`Schema`, defaults to `None`):
733
+ Schema for the created table. If not passed, names must be passed.
734
+ metadata (`Union[dict, Mapping]`, defaults to `None`):
735
+ Optional metadata for the schema (if inferred).
736
+
737
+ Returns:
738
+ `datasets.table.Table`
739
+ """
740
+ return cls(pa.Table.from_arrays(*args, **kwargs))
741
+
742
+ @classmethod
743
+ def from_pydict(cls, *args, **kwargs):
744
+ """
745
+ Construct a Table from Arrow arrays or columns.
746
+
747
+ Args:
748
+ mapping (`Union[dict, Mapping]`):
749
+ A mapping of strings to Arrays or Python lists.
750
+ schema (`Schema`, defaults to `None`):
751
+ If not passed, will be inferred from the Mapping values
752
+ metadata (`Union[dict, Mapping]`, defaults to `None`):
753
+ Optional metadata for the schema (if inferred).
754
+
755
+ Returns:
756
+ `datasets.table.Table`
757
+ """
758
+ return cls(pa.Table.from_pydict(*args, **kwargs))
759
+
760
+ @classmethod
761
+ def from_pylist(cls, mapping, *args, **kwargs):
762
+ """
763
+ Construct a Table from list of rows / dictionaries.
764
+
765
+ Args:
766
+ mapping (`List[dict]`):
767
+ A mapping of strings to row values.
768
+ schema (`Schema`, defaults to `None`):
769
+ If not passed, will be inferred from the Mapping values
770
+ metadata (`Union[dict, Mapping]`, defaults to `None`):
771
+ Optional metadata for the schema (if inferred).
772
+
773
+ Returns:
774
+ `datasets.table.Table`
775
+ """
776
+ return cls(pa.Table.from_pylist(mapping, *args, **kwargs))
777
+
778
+ @classmethod
779
+ def from_batches(cls, *args, **kwargs):
780
+ """
781
+ Construct a Table from a sequence or iterator of Arrow `RecordBatches`.
782
+
783
+ Args:
784
+ batches (`Union[Sequence[pyarrow.RecordBatch], Iterator[pyarrow.RecordBatch]]`):
785
+ Sequence of `RecordBatch` to be converted, all schemas must be equal.
786
+ schema (`Schema`, defaults to `None`):
787
+ If not passed, will be inferred from the first `RecordBatch`.
788
+
789
+ Returns:
790
+ `datasets.table.Table`:
791
+ """
792
+ return cls(pa.Table.from_batches(*args, **kwargs))
793
+
794
+ def slice(self, offset=0, length=None):
795
+ """
796
+ Compute zero-copy slice of this Table.
797
+
798
+ Args:
799
+ offset (`int`, defaults to `0`):
800
+ Offset from start of table to slice.
801
+ length (`int`, defaults to `None`):
802
+ Length of slice (default is until end of table starting from
803
+ offset).
804
+
805
+ Returns:
806
+ `datasets.table.Table`
807
+ """
808
+ # Use fast slicing here
809
+ return InMemoryTable(self.fast_slice(offset=offset, length=length))
810
+
811
+ def filter(self, *args, **kwargs):
812
+ """
813
+ Select records from a Table. See `pyarrow.compute.filter` for full usage.
814
+ """
815
+ return InMemoryTable(self.table.filter(*args, **kwargs))
816
+
817
+ def flatten(self, *args, **kwargs):
818
+ """
819
+ Flatten this Table. Each column with a struct type is flattened
820
+ into one column per struct field. Other columns are left unchanged.
821
+
822
+ Args:
823
+ memory_pool (`MemoryPool`, defaults to `None`):
824
+ For memory allocations, if required, otherwise use default pool.
825
+
826
+ Returns:
827
+ `datasets.table.Table`
828
+ """
829
+ return InMemoryTable(table_flatten(self.table, *args, **kwargs))
830
+
831
+ def combine_chunks(self, *args, **kwargs):
832
+ """
833
+ Make a new table by combining the chunks this table has.
834
+
835
+ All the underlying chunks in the `ChunkedArray` of each column are
836
+ concatenated into zero or one chunk.
837
+
838
+ Args:
839
+ memory_pool (`MemoryPool`, defaults to `None`):
840
+ For memory allocations, if required, otherwise use default pool.
841
+
842
+ Returns:
843
+ `datasets.table.Table`
844
+ """
845
+ return InMemoryTable(self.table.combine_chunks(*args, **kwargs))
846
+
847
+ def cast(self, *args, **kwargs):
848
+ """
849
+ Cast table values to another schema.
850
+
851
+ Args:
852
+ target_schema (`Schema`):
853
+ Schema to cast to, the names and order of fields must match.
854
+ safe (`bool`, defaults to `True`):
855
+ Check for overflows or other unsafe conversions.
856
+
857
+ Returns:
858
+ `datasets.table.Table`
859
+ """
860
+ return InMemoryTable(table_cast(self.table, *args, **kwargs))
861
+
862
+ def replace_schema_metadata(self, *args, **kwargs):
863
+ """
864
+ EXPERIMENTAL: Create shallow copy of table by replacing schema
865
+ key-value metadata with the indicated new metadata (which may be `None`,
866
+ which deletes any existing metadata).
867
+
868
+ Args:
869
+ metadata (`dict`, defaults to `None`):
870
+
871
+ Returns:
872
+ `datasets.table.Table`: shallow_copy
873
+ """
874
+ return InMemoryTable(self.table.replace_schema_metadata(*args, **kwargs))
875
+
876
+ def add_column(self, *args, **kwargs):
877
+ """
878
+ Add column to Table at position.
879
+
880
+ A new table is returned with the column added, the original table
881
+ object is left unchanged.
882
+
883
+ Args:
884
+ i (`int`):
885
+ Index to place the column at.
886
+ field_ (`Union[str, pyarrow.Field]`):
887
+ If a string is passed then the type is deduced from the column
888
+ data.
889
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
890
+ Column data.
891
+
892
+ Returns:
893
+ `datasets.table.Table`: New table with the passed column added.
894
+ """
895
+ return InMemoryTable(self.table.add_column(*args, **kwargs))
896
+
897
+ def append_column(self, *args, **kwargs):
898
+ """
899
+ Append column at end of columns.
900
+
901
+ Args:
902
+ field_ (`Union[str, pyarrow.Field]`):
903
+ If a string is passed then the type is deduced from the column
904
+ data.
905
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
906
+ Column data.
907
+
908
+ Returns:
909
+ `datasets.table.Table`:
910
+ New table with the passed column added.
911
+ """
912
+ return InMemoryTable(self.table.append_column(*args, **kwargs))
913
+
914
+ def remove_column(self, *args, **kwargs):
915
+ """
916
+ Create new Table with the indicated column removed.
917
+
918
+ Args:
919
+ i (`int`):
920
+ Index of column to remove.
921
+
922
+ Returns:
923
+ `datasets.table.Table`:
924
+ New table without the column.
925
+ """
926
+ return InMemoryTable(self.table.remove_column(*args, **kwargs))
927
+
928
+ def set_column(self, *args, **kwargs):
929
+ """
930
+ Replace column in Table at position.
931
+
932
+ Args:
933
+ i (`int`):
934
+ Index to place the column at.
935
+ field_ (`Union[str, pyarrow.Field]`):
936
+ If a string is passed then the type is deduced from the column
937
+ data.
938
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
939
+ Column data.
940
+
941
+ Returns:
942
+ `datasets.table.Table`:
943
+ New table with the passed column set.
944
+ """
945
+ return InMemoryTable(self.table.set_column(*args, **kwargs))
946
+
947
+ def rename_columns(self, *args, **kwargs):
948
+ """
949
+ Create new table with columns renamed to provided names.
950
+ """
951
+ return InMemoryTable(self.table.rename_columns(*args, **kwargs))
952
+
953
+ def drop(self, *args, **kwargs):
954
+ """
955
+ Drop one or more columns and return a new table.
956
+
957
+ Args:
958
+ columns (`List[str]`):
959
+ List of field names referencing existing columns.
960
+
961
+ Raises:
962
+ `KeyError` : if any of the passed columns name are not existing.
963
+
964
+ Returns:
965
+ `datasets.table.Table`:
966
+ New table without the columns.
967
+ """
968
+ return InMemoryTable(self.table.drop(*args, **kwargs))
969
+
970
+ def select(self, *args, **kwargs):
971
+ """
972
+ Select columns of the table.
973
+
974
+ Returns a new table with the specified columns, and metadata preserved.
975
+
976
+ Args:
977
+ columns (:obj:`Union[List[str], List[int]]`):
978
+ The column names or integer indices to select.
979
+
980
+ Returns:
981
+ :class:`datasets.table.Table`: New table with the specified columns, and metadata preserved.
982
+ """
983
+ return InMemoryTable(self.table.select(*args, **kwargs))
984
+
985
+
986
+ # The MemoryMappedTable needs replays to properly reload tables from the disk
987
+ Replay = Tuple[str, tuple, dict]
988
+
989
+
990
+ class MemoryMappedTable(TableBlock):
991
+ """
992
+ The table is said memory mapped when it doesn't use the user's RAM but loads the data
993
+ from the disk instead.
994
+
995
+ Pickling it doesn't copy the data into memory.
996
+ Instead, only the path to the memory mapped arrow file is pickled, as well as the list
997
+ of transforms to "replay" when reloading the table from the disk.
998
+
999
+ Its implementation requires to store an history of all the transforms that were applied
1000
+ to the underlying pyarrow Table, so that they can be "replayed" when reloading the Table
1001
+ from the disk.
1002
+
1003
+ This is different from the `InMemoryTable` table, for which pickling does copy all the
1004
+ data in memory.
1005
+
1006
+ `InMemoryTable` must be used when data fit in memory, while `MemoryMapped` are reserved for
1007
+ data bigger than memory or when you want the memory footprint of your application to
1008
+ stay low.
1009
+ """
1010
+
1011
+ def __init__(self, table: pa.Table, path: str, replays: Optional[List[Replay]] = None):
1012
+ super().__init__(table)
1013
+ self.path = os.path.abspath(path)
1014
+ self.replays: List[Replay] = replays if replays is not None else []
1015
+
1016
+ @classmethod
1017
+ def from_file(cls, filename: str, replays=None):
1018
+ table = _memory_mapped_arrow_table_from_file(filename)
1019
+ table = cls._apply_replays(table, replays)
1020
+ return cls(table, filename, replays)
1021
+
1022
+ def __getstate__(self):
1023
+ return {"path": self.path, "replays": self.replays}
1024
+
1025
+ def __setstate__(self, state):
1026
+ path = state["path"]
1027
+ replays = state["replays"]
1028
+ table = _memory_mapped_arrow_table_from_file(path)
1029
+ table = self._apply_replays(table, replays)
1030
+ MemoryMappedTable.__init__(self, table, path=path, replays=replays)
1031
+
1032
+ @staticmethod
1033
+ def _apply_replays(table: pa.Table, replays: Optional[List[Replay]] = None) -> pa.Table:
1034
+ if replays is not None:
1035
+ for name, args, kwargs in replays:
1036
+ if name == "cast":
1037
+ table = table_cast(table, *args, **kwargs)
1038
+ elif name == "flatten":
1039
+ table = table_flatten(table, *args, **kwargs)
1040
+ else:
1041
+ table = getattr(table, name)(*args, **kwargs)
1042
+ return table
1043
+
1044
+ def _append_replay(self, replay: Replay) -> List[Replay]:
1045
+ replays = copy.deepcopy(self.replays)
1046
+ replays.append(replay)
1047
+ return replays
1048
+
1049
+ def slice(self, offset=0, length=None):
1050
+ """
1051
+ Compute zero-copy slice of this Table.
1052
+
1053
+ Args:
1054
+ offset (`int`, defaults to `0`):
1055
+ Offset from start of table to slice.
1056
+ length (`int`, defaults to `None`):
1057
+ Length of slice (default is until end of table starting from
1058
+ offset).
1059
+
1060
+ Returns:
1061
+ `datasets.table.Table`
1062
+ """
1063
+ replay = ("slice", (offset, length), {})
1064
+ replays = self._append_replay(replay)
1065
+ # Use fast slicing here
1066
+ return MemoryMappedTable(self.fast_slice(offset=offset, length=length), self.path, replays)
1067
+
1068
+ def filter(self, *args, **kwargs):
1069
+ """
1070
+ Select records from a Table. See `pyarrow.compute.filter` for full usage.
1071
+ """
1072
+ replay = ("filter", copy.deepcopy(args), copy.deepcopy(kwargs))
1073
+ replays = self._append_replay(replay)
1074
+ return MemoryMappedTable(self.table.filter(*args, **kwargs), self.path, replays)
1075
+
1076
+ def flatten(self, *args, **kwargs):
1077
+ """
1078
+ Flatten this Table. Each column with a struct type is flattened
1079
+ into one column per struct field. Other columns are left unchanged.
1080
+
1081
+ Args:
1082
+ memory_pool (`MemoryPool`, defaults to `None`):
1083
+ For memory allocations, if required, otherwise use default pool.
1084
+
1085
+ Returns:
1086
+ `datasets.table.Table`
1087
+ """
1088
+ replay = ("flatten", copy.deepcopy(args), copy.deepcopy(kwargs))
1089
+ replays = self._append_replay(replay)
1090
+ return MemoryMappedTable(table_flatten(self.table, *args, **kwargs), self.path, replays)
1091
+
1092
+ def combine_chunks(self, *args, **kwargs):
1093
+ """
1094
+ Make a new table by combining the chunks this table has.
1095
+
1096
+ All the underlying chunks in the ChunkedArray of each column are
1097
+ concatenated into zero or one chunk.
1098
+
1099
+ Args:
1100
+ memory_pool (`MemoryPool`, defaults to `None`):
1101
+ For memory allocations, if required, otherwise use default pool.
1102
+
1103
+ Returns:
1104
+ `datasets.table.Table`
1105
+ """
1106
+ replay = ("combine_chunks", copy.deepcopy(args), copy.deepcopy(kwargs))
1107
+ replays = self._append_replay(replay)
1108
+ return MemoryMappedTable(self.table.combine_chunks(*args, **kwargs), self.path, replays)
1109
+
1110
+ def cast(self, *args, **kwargs):
1111
+ """
1112
+ Cast table values to another schema
1113
+
1114
+ Args:
1115
+ target_schema (`Schema`):
1116
+ Schema to cast to, the names and order of fields must match.
1117
+ safe (`bool`, defaults to `True`):
1118
+ Check for overflows or other unsafe conversions.
1119
+
1120
+ Returns:
1121
+ `datasets.table.Table`
1122
+ """
1123
+ replay = ("cast", copy.deepcopy(args), copy.deepcopy(kwargs))
1124
+ replays = self._append_replay(replay)
1125
+ return MemoryMappedTable(table_cast(self.table, *args, **kwargs), self.path, replays)
1126
+
1127
+ def replace_schema_metadata(self, *args, **kwargs):
1128
+ """
1129
+ EXPERIMENTAL: Create shallow copy of table by replacing schema
1130
+ key-value metadata with the indicated new metadata (which may be None,
1131
+ which deletes any existing metadata.
1132
+
1133
+ Args:
1134
+ metadata (`dict`, defaults to `None`):
1135
+
1136
+ Returns:
1137
+ `datasets.table.Table`: shallow_copy
1138
+ """
1139
+ replay = ("replace_schema_metadata", copy.deepcopy(args), copy.deepcopy(kwargs))
1140
+ replays = self._append_replay(replay)
1141
+ return MemoryMappedTable(self.table.replace_schema_metadata(*args, **kwargs), self.path, replays)
1142
+
1143
+ def add_column(self, *args, **kwargs):
1144
+ """
1145
+ Add column to Table at position.
1146
+
1147
+ A new table is returned with the column added, the original table
1148
+ object is left unchanged.
1149
+
1150
+ Args:
1151
+ i (`int`):
1152
+ Index to place the column at.
1153
+ field_ (`Union[str, pyarrow.Field]`):
1154
+ If a string is passed then the type is deduced from the column
1155
+ data.
1156
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
1157
+ Column data.
1158
+
1159
+ Returns:
1160
+ `datasets.table.Table`: New table with the passed column added.
1161
+ """
1162
+ replay = ("add_column", copy.deepcopy(args), copy.deepcopy(kwargs))
1163
+ replays = self._append_replay(replay)
1164
+ return MemoryMappedTable(self.table.add_column(*args, **kwargs), self.path, replays)
1165
+
1166
+ def append_column(self, *args, **kwargs):
1167
+ """
1168
+ Append column at end of columns.
1169
+
1170
+ Args:
1171
+ field_ (`Union[str, pyarrow.Field]`):
1172
+ If a string is passed then the type is deduced from the column
1173
+ data.
1174
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
1175
+ Column data.
1176
+
1177
+ Returns:
1178
+ `datasets.table.Table`:
1179
+ New table with the passed column added.
1180
+ """
1181
+ replay = ("append_column", copy.deepcopy(args), copy.deepcopy(kwargs))
1182
+ replays = self._append_replay(replay)
1183
+ return MemoryMappedTable(self.table.append_column(*args, **kwargs), self.path, replays)
1184
+
1185
+ def remove_column(self, *args, **kwargs):
1186
+ """
1187
+ Create new Table with the indicated column removed.
1188
+
1189
+ Args:
1190
+ i (`int`):
1191
+ Index of column to remove.
1192
+
1193
+ Returns:
1194
+ `datasets.table.Table`:
1195
+ New table without the column.
1196
+ """
1197
+ replay = ("remove_column", copy.deepcopy(args), copy.deepcopy(kwargs))
1198
+ replays = self._append_replay(replay)
1199
+ return MemoryMappedTable(self.table.remove_column(*args, **kwargs), self.path, replays)
1200
+
1201
+ def set_column(self, *args, **kwargs):
1202
+ """
1203
+ Replace column in Table at position.
1204
+
1205
+ Args:
1206
+ i (`int`):
1207
+ Index to place the column at.
1208
+ field_ (`Union[str, pyarrow.Field]`):
1209
+ If a string is passed then the type is deduced from the column
1210
+ data.
1211
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
1212
+ Column data.
1213
+
1214
+ Returns:
1215
+ `datasets.table.Table`:
1216
+ New table with the passed column set.
1217
+ """
1218
+ replay = ("set_column", copy.deepcopy(args), copy.deepcopy(kwargs))
1219
+ replays = self._append_replay(replay)
1220
+ return MemoryMappedTable(self.table.set_column(*args, **kwargs), self.path, replays)
1221
+
1222
+ def rename_columns(self, *args, **kwargs):
1223
+ """
1224
+ Create new table with columns renamed to provided names.
1225
+ """
1226
+ replay = ("rename_columns", copy.deepcopy(args), copy.deepcopy(kwargs))
1227
+ replays = self._append_replay(replay)
1228
+ return MemoryMappedTable(self.table.rename_columns(*args, **kwargs), self.path, replays)
1229
+
1230
+ def drop(self, *args, **kwargs):
1231
+ """
1232
+ Drop one or more columns and return a new table.
1233
+
1234
+ Args:
1235
+ columns (`List[str]`):
1236
+ List of field names referencing existing columns.
1237
+
1238
+ Raises:
1239
+ `KeyError` : if any of the passed columns name are not existing.
1240
+
1241
+ Returns:
1242
+ `datasets.table.Table`:
1243
+ New table without the columns.
1244
+ """
1245
+ replay = ("drop", copy.deepcopy(args), copy.deepcopy(kwargs))
1246
+ replays = self._append_replay(replay)
1247
+ return MemoryMappedTable(self.table.drop(*args, **kwargs), self.path, replays)
1248
+
1249
+ def select(self, *args, **kwargs):
1250
+ """
1251
+ Select columns of the table.
1252
+
1253
+ Returns a new table with the specified columns, and metadata preserved.
1254
+
1255
+ Args:
1256
+ columns (:obj:`Union[List[str], List[int]]`):
1257
+ The column names or integer indices to select.
1258
+
1259
+ Returns:
1260
+ :class:`datasets.table.Table`: New table with the specified columns, and metadata preserved.
1261
+ """
1262
+ replay = ("select", copy.deepcopy(args), copy.deepcopy(kwargs))
1263
+ replays = self._append_replay(replay)
1264
+ return MemoryMappedTable(self.table.select(*args, **kwargs), self.path, replays)
1265
+
1266
+
1267
+ # A ConcatenationTable is the concatenation of several tables.
1268
+ # The ``blocks`` attributes stores a list of list of blocks.
1269
+ # The first axis concatenates the tables along the axis 0 (it appends rows),
1270
+ # while the second axis concatenates tables along the axis 1 (it appends columns).
1271
+ TableBlockContainer = TypeVar("TableBlockContainer", TableBlock, List[TableBlock], List[List[TableBlock]])
1272
+
1273
+
1274
+ class ConcatenationTable(Table):
1275
+ """
1276
+ The table comes from the concatenation of several tables called blocks.
1277
+ It enables concatenation on both axis 0 (append rows) and axis 1 (append columns).
1278
+
1279
+ The underlying tables are called "blocks" and can be either `InMemoryTable`
1280
+ or `MemoryMappedTable` objects.
1281
+ This allows to combine tables that come from memory or that are memory mapped.
1282
+ When a `ConcatenationTable` is pickled, then each block is pickled:
1283
+ - the `InMemoryTable` objects are pickled by copying all the data in memory.
1284
+ - the MemoryMappedTable objects are pickled without copying the data into memory.
1285
+ Instead, only the path to the memory mapped arrow file is pickled, as well as the list
1286
+ of transforms to "replays" when reloading the table from the disk.
1287
+
1288
+ Its implementation requires to store each block separately.
1289
+ The `blocks` attributes stores a list of list of blocks.
1290
+ The first axis concatenates the tables along the axis 0 (it appends rows),
1291
+ while the second axis concatenates tables along the axis 1 (it appends columns).
1292
+
1293
+ If some columns are missing when concatenating on axis 0, they are filled with null values.
1294
+ This is done using `pyarrow.concat_tables(tables, promote=True)`.
1295
+
1296
+ You can access the fully combined table by accessing the `ConcatenationTable.table` attribute,
1297
+ and the blocks by accessing the `ConcatenationTable.blocks` attribute.
1298
+ """
1299
+
1300
+ def __init__(self, table: pa.Table, blocks: List[List[TableBlock]]):
1301
+ super().__init__(table)
1302
+ self.blocks = blocks
1303
+ # Check that all the blocks have the right type.
1304
+ # Only InMemoryTable and MemoryMappedTable are allowed.
1305
+ for subtables in blocks:
1306
+ for subtable in subtables:
1307
+ if not isinstance(subtable, TableBlock):
1308
+ raise TypeError(
1309
+ "The blocks of a ConcatenationTable must be InMemoryTable or MemoryMappedTable objects"
1310
+ f", but got {subtable}."
1311
+ )
1312
+
1313
+ def __getstate__(self):
1314
+ return {"blocks": self.blocks, "schema": self.table.schema}
1315
+
1316
+ def __setstate__(self, state):
1317
+ blocks = state["blocks"]
1318
+ schema = state["schema"]
1319
+ table = self._concat_blocks_horizontally_and_vertically(blocks)
1320
+ if schema is not None and table.schema != schema:
1321
+ # We fix the columns by concatenating with an empty table with the right columns
1322
+ empty_table = pa.Table.from_batches([], schema=schema)
1323
+ # we set promote=True to fill missing columns with null values
1324
+ if config.PYARROW_VERSION.major < 14:
1325
+ table = pa.concat_tables([table, empty_table], promote=True)
1326
+ else:
1327
+ table = pa.concat_tables([table, empty_table], promote_options="default")
1328
+ ConcatenationTable.__init__(self, table, blocks=blocks)
1329
+
1330
+ @staticmethod
1331
+ def _concat_blocks(blocks: List[Union[TableBlock, pa.Table]], axis: int = 0) -> pa.Table:
1332
+ pa_tables = [table.table if hasattr(table, "table") else table for table in blocks]
1333
+ if axis == 0:
1334
+ # we set promote=True to fill missing columns with null values
1335
+ if config.PYARROW_VERSION.major < 14:
1336
+ return pa.concat_tables(pa_tables, promote=True)
1337
+ else:
1338
+ return pa.concat_tables(pa_tables, promote_options="default")
1339
+ elif axis == 1:
1340
+ for i, table in enumerate(pa_tables):
1341
+ if i == 0:
1342
+ pa_table = table
1343
+ else:
1344
+ for name, col in zip(table.column_names, table.columns):
1345
+ pa_table = pa_table.append_column(name, col)
1346
+ return pa_table
1347
+ else:
1348
+ raise ValueError("'axis' must be either 0 or 1")
1349
+
1350
+ @classmethod
1351
+ def _concat_blocks_horizontally_and_vertically(cls, blocks: List[List[TableBlock]]) -> pa.Table:
1352
+ pa_tables_to_concat_vertically = []
1353
+ for i, tables in enumerate(blocks):
1354
+ if not tables:
1355
+ continue
1356
+ pa_table_horizontally_concatenated = cls._concat_blocks(tables, axis=1)
1357
+ pa_tables_to_concat_vertically.append(pa_table_horizontally_concatenated)
1358
+ return cls._concat_blocks(pa_tables_to_concat_vertically, axis=0)
1359
+
1360
+ @classmethod
1361
+ def _merge_blocks(cls, blocks: TableBlockContainer, axis: Optional[int] = None) -> TableBlockContainer:
1362
+ if axis is not None:
1363
+ merged_blocks = []
1364
+ for is_in_memory, block_group in groupby(blocks, key=lambda x: isinstance(x, InMemoryTable)):
1365
+ if is_in_memory:
1366
+ block_group = [InMemoryTable(cls._concat_blocks(list(block_group), axis=axis))]
1367
+ merged_blocks += list(block_group)
1368
+ else: # both
1369
+ merged_blocks = [cls._merge_blocks(row_block, axis=1) for row_block in blocks]
1370
+ if all(len(row_block) == 1 for row_block in merged_blocks):
1371
+ merged_blocks = cls._merge_blocks(
1372
+ [block for row_block in merged_blocks for block in row_block], axis=0
1373
+ )
1374
+ return merged_blocks
1375
+
1376
+ @classmethod
1377
+ def _consolidate_blocks(cls, blocks: TableBlockContainer) -> TableBlockContainer:
1378
+ if isinstance(blocks, TableBlock):
1379
+ return blocks
1380
+ elif isinstance(blocks[0], TableBlock):
1381
+ return cls._merge_blocks(blocks, axis=0)
1382
+ else:
1383
+ return cls._merge_blocks(blocks)
1384
+
1385
+ @classmethod
1386
+ def from_blocks(cls, blocks: TableBlockContainer) -> "ConcatenationTable":
1387
+ blocks = cls._consolidate_blocks(blocks)
1388
+ if isinstance(blocks, TableBlock):
1389
+ table = blocks
1390
+ return cls(table.table, [[table]])
1391
+ elif isinstance(blocks[0], TableBlock):
1392
+ table = cls._concat_blocks(blocks, axis=0)
1393
+ blocks = [[t] for t in blocks]
1394
+ return cls(table, blocks)
1395
+ else:
1396
+ table = cls._concat_blocks_horizontally_and_vertically(blocks)
1397
+ return cls(table, blocks)
1398
+
1399
+ @classmethod
1400
+ def from_tables(cls, tables: List[Union[pa.Table, Table]], axis: int = 0) -> "ConcatenationTable":
1401
+ """Create `ConcatenationTable` from list of tables.
1402
+
1403
+ Args:
1404
+ tables (list of `Table` or list of `pyarrow.Table`):
1405
+ List of tables.
1406
+ axis (`{0, 1}`, defaults to `0`, meaning over rows):
1407
+ Axis to concatenate over, where `0` means over rows (vertically) and `1` means over columns
1408
+ (horizontally).
1409
+
1410
+ <Added version="1.6.0"/>
1411
+ """
1412
+
1413
+ def to_blocks(table: Union[pa.Table, Table]) -> List[List[TableBlock]]:
1414
+ if isinstance(table, pa.Table):
1415
+ return [[InMemoryTable(table)]]
1416
+ elif isinstance(table, ConcatenationTable):
1417
+ return copy.deepcopy(table.blocks)
1418
+ else:
1419
+ return [[table]]
1420
+
1421
+ def _slice_row_block(row_block: List[TableBlock], length: int) -> Tuple[List[TableBlock], List[TableBlock]]:
1422
+ sliced = [table.slice(0, length) for table in row_block]
1423
+ remainder = [table.slice(length, len(row_block[0]) - length) for table in row_block]
1424
+ return sliced, remainder
1425
+
1426
+ def _split_both_like(
1427
+ result: List[List[TableBlock]], blocks: List[List[TableBlock]]
1428
+ ) -> Tuple[List[List[TableBlock]], List[List[TableBlock]]]:
1429
+ """
1430
+ Make sure each row_block contain the same num_rows to be able to concatenate them on axis=1.
1431
+
1432
+ To do so, we modify both blocks sets to have the same row_blocks boundaries.
1433
+ For example, if `result` has 2 row_blocks of 3 rows and `blocks` has 3 row_blocks of 2 rows,
1434
+ we modify both to have 4 row_blocks of size 2, 1, 1 and 2:
1435
+
1436
+ [ x x x | x x x ]
1437
+ + [ y y | y y | y y ]
1438
+ -----------------------------
1439
+ = [ x x | x | x | x x ]
1440
+ [ y y | y | y | y y ]
1441
+
1442
+ """
1443
+ result, blocks = list(result), list(blocks)
1444
+ new_result, new_blocks = [], []
1445
+ while result and blocks:
1446
+ # we slice the longest row block to save two row blocks of same length
1447
+ # and we replace the long row block by its remainder if necessary
1448
+ if len(result[0][0]) > len(blocks[0][0]):
1449
+ new_blocks.append(blocks[0])
1450
+ sliced, result[0] = _slice_row_block(result[0], len(blocks.pop(0)[0]))
1451
+ new_result.append(sliced)
1452
+ elif len(result[0][0]) < len(blocks[0][0]):
1453
+ new_result.append(result[0])
1454
+ sliced, blocks[0] = _slice_row_block(blocks[0], len(result.pop(0)[0]))
1455
+ new_blocks.append(sliced)
1456
+ else:
1457
+ new_result.append(result.pop(0))
1458
+ new_blocks.append(blocks.pop(0))
1459
+ if result or blocks:
1460
+ raise ValueError("Failed to concatenate on axis=1 because tables don't have the same number of rows")
1461
+ return new_result, new_blocks
1462
+
1463
+ def _extend_blocks(
1464
+ result: List[List[TableBlock]], blocks: List[List[TableBlock]], axis: int = 0
1465
+ ) -> List[List[TableBlock]]:
1466
+ if axis == 0:
1467
+ result.extend(blocks)
1468
+ elif axis == 1:
1469
+ # We make sure each row_block have the same num_rows
1470
+ result, blocks = _split_both_like(result, blocks)
1471
+ for i, row_block in enumerate(blocks):
1472
+ result[i].extend(row_block)
1473
+ return result
1474
+
1475
+ blocks = to_blocks(tables[0])
1476
+ for table in tables[1:]:
1477
+ table_blocks = to_blocks(table)
1478
+ blocks = _extend_blocks(blocks, table_blocks, axis=axis)
1479
+ return cls.from_blocks(blocks)
1480
+
1481
+ @property
1482
+ def _slices(self):
1483
+ offset = 0
1484
+ for tables in self.blocks:
1485
+ length = len(tables[0])
1486
+ yield (offset, length)
1487
+ offset += length
1488
+
1489
+ def slice(self, offset=0, length=None):
1490
+ """
1491
+ Compute zero-copy slice of this Table.
1492
+
1493
+ Args:
1494
+ offset (`int`, defaults to `0`):
1495
+ Offset from start of table to slice.
1496
+ length (`int`, defaults to `None`):
1497
+ Length of slice (default is until end of table starting from
1498
+ offset).
1499
+
1500
+ Returns:
1501
+ `datasets.table.Table`
1502
+ """
1503
+ table = self.table.slice(offset, length=length)
1504
+ length = length if length is not None else self.num_rows - offset
1505
+ blocks = []
1506
+ for tables in self.blocks:
1507
+ n_rows = len(tables[0])
1508
+ if length == 0:
1509
+ break
1510
+ elif n_rows <= offset:
1511
+ offset = offset - n_rows
1512
+ elif n_rows <= offset + length:
1513
+ blocks.append([t.slice(offset) for t in tables])
1514
+ length, offset = length + offset - n_rows, 0
1515
+ else:
1516
+ blocks.append([t.slice(offset, length) for t in tables])
1517
+ length, offset = 0, 0
1518
+ return ConcatenationTable(table, blocks)
1519
+
1520
+ def filter(self, mask, *args, **kwargs):
1521
+ """
1522
+ Select records from a Table. See `pyarrow.compute.filter` for full usage.
1523
+ """
1524
+ table = self.table.filter(mask, *args, **kwargs)
1525
+ blocks = []
1526
+ for (offset, length), tables in zip(self._slices, self.blocks):
1527
+ submask = mask.slice(offset, length)
1528
+ blocks.append([t.filter(submask, *args, **kwargs) for t in tables])
1529
+ return ConcatenationTable(table, blocks)
1530
+
1531
+ def flatten(self, *args, **kwargs):
1532
+ """
1533
+ Flatten this Table. Each column with a struct type is flattened
1534
+ into one column per struct field. Other columns are left unchanged.
1535
+
1536
+ Args:
1537
+ memory_pool (`MemoryPool`, defaults to `None`):
1538
+ For memory allocations, if required, otherwise use default pool.
1539
+
1540
+ Returns:
1541
+ `datasets.table.Table`
1542
+ """
1543
+ table = table_flatten(self.table, *args, **kwargs)
1544
+ blocks = []
1545
+ for tables in self.blocks:
1546
+ blocks.append([t.flatten(*args, **kwargs) for t in tables])
1547
+ return ConcatenationTable(table, blocks)
1548
+
1549
+ def combine_chunks(self, *args, **kwargs):
1550
+ """
1551
+ Make a new table by combining the chunks this table has.
1552
+
1553
+ All the underlying chunks in the `ChunkedArray` of each column are
1554
+ concatenated into zero or one chunk.
1555
+
1556
+ Args:
1557
+ memory_pool (`MemoryPool`, defaults to `None`):
1558
+ For memory allocations, if required, otherwise use default pool.
1559
+
1560
+ Returns:
1561
+ `datasets.table.Table`
1562
+ """
1563
+ table = self.table.combine_chunks(*args, **kwargs)
1564
+ blocks = []
1565
+ for tables in self.blocks:
1566
+ blocks.append([t.combine_chunks(*args, **kwargs) for t in tables])
1567
+ return ConcatenationTable(table, blocks)
1568
+
1569
+ def cast(self, target_schema, *args, **kwargs):
1570
+ """
1571
+ Cast table values to another schema.
1572
+
1573
+ Args:
1574
+ target_schema (`Schema`):
1575
+ Schema to cast to, the names and order of fields must match.
1576
+ safe (`bool`, defaults to `True`):
1577
+ Check for overflows or other unsafe conversions.
1578
+
1579
+ Returns:
1580
+ `datasets.table.Table`
1581
+ """
1582
+ from .features import Features
1583
+
1584
+ table = table_cast(self.table, target_schema, *args, **kwargs)
1585
+ target_features = Features.from_arrow_schema(target_schema)
1586
+ blocks = []
1587
+ for subtables in self.blocks:
1588
+ new_tables = []
1589
+ fields = list(target_schema)
1590
+ for subtable in subtables:
1591
+ subfields = []
1592
+ for name in subtable.column_names:
1593
+ subfields.append(fields.pop(next(i for i, field in enumerate(fields) if field.name == name)))
1594
+ subfeatures = Features({subfield.name: target_features[subfield.name] for subfield in subfields})
1595
+ subschema = subfeatures.arrow_schema
1596
+ new_tables.append(subtable.cast(subschema, *args, **kwargs))
1597
+ blocks.append(new_tables)
1598
+ return ConcatenationTable(table, blocks)
1599
+
1600
+ def replace_schema_metadata(self, *args, **kwargs):
1601
+ """
1602
+ EXPERIMENTAL: Create shallow copy of table by replacing schema
1603
+ key-value metadata with the indicated new metadata (which may be `None`,
1604
+ which deletes any existing metadata).
1605
+
1606
+ Args:
1607
+ metadata (`dict`, defaults to `None`):
1608
+
1609
+ Returns:
1610
+ `datasets.table.Table`: shallow_copy
1611
+ """
1612
+ table = self.table.replace_schema_metadata(*args, **kwargs)
1613
+ blocks = []
1614
+ for tables in self.blocks:
1615
+ blocks.append([t.replace_schema_metadata(*args, **kwargs) for t in tables])
1616
+ return ConcatenationTable(table, self.blocks)
1617
+
1618
+ def add_column(self, *args, **kwargs):
1619
+ """
1620
+ Add column to Table at position.
1621
+
1622
+ A new table is returned with the column added, the original table
1623
+ object is left unchanged.
1624
+
1625
+ Args:
1626
+ i (`int`):
1627
+ Index to place the column at.
1628
+ field_ (`Union[str, pyarrow.Field]`):
1629
+ If a string is passed then the type is deduced from the column
1630
+ data.
1631
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
1632
+ Column data.
1633
+
1634
+ Returns:
1635
+ `datasets.table.Table`: New table with the passed column added.
1636
+ """
1637
+ raise NotImplementedError()
1638
+
1639
+ def append_column(self, *args, **kwargs):
1640
+ """
1641
+ Append column at end of columns.
1642
+
1643
+ Args:
1644
+ field_ (`Union[str, pyarrow.Field]`):
1645
+ If a string is passed then the type is deduced from the column
1646
+ data.
1647
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
1648
+ Column data.
1649
+
1650
+ Returns:
1651
+ `datasets.table.Table`:
1652
+ New table with the passed column added.
1653
+ """
1654
+ raise NotImplementedError()
1655
+
1656
+ def remove_column(self, i, *args, **kwargs):
1657
+ """
1658
+ Create new Table with the indicated column removed.
1659
+
1660
+ Args:
1661
+ i (`int`):
1662
+ Index of column to remove.
1663
+
1664
+ Returns:
1665
+ `datasets.table.Table`:
1666
+ New table without the column.
1667
+ """
1668
+ table = self.table.remove_column(i, *args, **kwargs)
1669
+ name = self.table.column_names[i]
1670
+ blocks = []
1671
+ for tables in self.blocks:
1672
+ blocks.append(
1673
+ [
1674
+ t.remove_column(t.column_names.index(name), *args, **kwargs) if name in t.column_names else t
1675
+ for t in tables
1676
+ ]
1677
+ )
1678
+ return ConcatenationTable(table, blocks)
1679
+
1680
+ def set_column(self, *args, **kwargs):
1681
+ """
1682
+ Replace column in Table at position.
1683
+
1684
+ Args:
1685
+ i (`int`):
1686
+ Index to place the column at.
1687
+ field_ (`Union[str, pyarrow.Field]`):
1688
+ If a string is passed then the type is deduced from the column
1689
+ data.
1690
+ column (`Union[pyarrow.Array, List[pyarrow.Array]]`):
1691
+ Column data.
1692
+
1693
+ Returns:
1694
+ `datasets.table.Table`:
1695
+ New table with the passed column set.
1696
+ """
1697
+ raise NotImplementedError()
1698
+
1699
+ def rename_columns(self, names, *args, **kwargs):
1700
+ """
1701
+ Create new table with columns renamed to provided names.
1702
+ """
1703
+ table = self.table.rename_columns(names, *args, **kwargs)
1704
+ names = dict(zip(self.table.column_names, names))
1705
+ blocks = []
1706
+ for tables in self.blocks:
1707
+ blocks.append(
1708
+ [t.rename_columns([names[name] for name in t.column_names], *args, **kwargs) for t in tables]
1709
+ )
1710
+ return ConcatenationTable(table, blocks)
1711
+
1712
+ def drop(self, columns, *args, **kwargs):
1713
+ """
1714
+ Drop one or more columns and return a new table.
1715
+
1716
+ Args:
1717
+ columns (`List[str]`):
1718
+ List of field names referencing existing columns.
1719
+
1720
+ Raises:
1721
+ `KeyError` : if any of the passed columns name are not existing.
1722
+
1723
+ Returns:
1724
+ `datasets.table.Table`:
1725
+ New table without the columns.
1726
+ """
1727
+ table = self.table.drop(columns, *args, **kwargs)
1728
+ blocks = []
1729
+ for tables in self.blocks:
1730
+ blocks.append([t.drop([c for c in columns if c in t.column_names], *args, **kwargs) for t in tables])
1731
+ return ConcatenationTable(table, blocks)
1732
+
1733
+ def select(self, columns, *args, **kwargs):
1734
+ """
1735
+ Select columns of the table.
1736
+
1737
+ Returns a new table with the specified columns, and metadata preserved.
1738
+
1739
+ Args:
1740
+ columns (:obj:`Union[List[str], List[int]]`):
1741
+ The column names or integer indices to select.
1742
+
1743
+ Returns:
1744
+ :class:`datasets.table.Table`: New table with the specified columns, and metadata preserved.
1745
+ """
1746
+ table = self.table.select(columns, *args, **kwargs)
1747
+ blocks = []
1748
+ for tables in self.blocks:
1749
+ blocks.append([t.select([c for c in columns if c in t.column_names], *args, **kwargs) for t in tables])
1750
+ return ConcatenationTable(table, blocks)
1751
+
1752
+
1753
+ def concat_tables(tables: List[Table], axis: int = 0) -> Table:
1754
+ """
1755
+ Concatenate tables.
1756
+
1757
+ Args:
1758
+ tables (list of `Table`):
1759
+ List of tables to be concatenated.
1760
+ axis (`{0, 1}`, defaults to `0`, meaning over rows):
1761
+ Axis to concatenate over, where `0` means over rows (vertically) and `1` means over columns
1762
+ (horizontally).
1763
+
1764
+ <Added version="1.6.0"/>
1765
+ Returns:
1766
+ `datasets.table.Table`:
1767
+ If the number of input tables is > 1, then the returned table is a `datasets.table.ConcatenationTable`.
1768
+ Otherwise if there's only one table, it is returned as is.
1769
+ """
1770
+ tables = list(tables)
1771
+ if len(tables) == 1:
1772
+ return tables[0]
1773
+ return ConcatenationTable.from_tables(tables, axis=axis)
1774
+
1775
+
1776
+ def list_table_cache_files(table: Table) -> List[str]:
1777
+ """
1778
+ Get the cache files that are loaded by the table.
1779
+ Cache file are used when parts of the table come from the disk via memory mapping.
1780
+
1781
+ Returns:
1782
+ `List[str]`:
1783
+ A list of paths to the cache files loaded by the table.
1784
+ """
1785
+ if isinstance(table, ConcatenationTable):
1786
+ cache_files = []
1787
+ for subtables in table.blocks:
1788
+ for subtable in subtables:
1789
+ cache_files += list_table_cache_files(subtable)
1790
+ return cache_files
1791
+ elif isinstance(table, MemoryMappedTable):
1792
+ return [table.path]
1793
+ else:
1794
+ return []
1795
+
1796
+
1797
+ def _wrap_for_chunked_arrays(func):
1798
+ """Apply the function on each chunk of a `pyarrow.ChunkedArray`, or on the array directly"""
1799
+
1800
+ def wrapper(array, *args, **kwargs):
1801
+ if isinstance(array, pa.ChunkedArray):
1802
+ return pa.chunked_array([func(chunk, *args, **kwargs) for chunk in array.chunks])
1803
+ else:
1804
+ return func(array, *args, **kwargs)
1805
+
1806
+ return wrapper
1807
+
1808
+
1809
+ def _are_list_values_of_length(array: pa.ListArray, length: int) -> bool:
1810
+ """Check if all the sub-lists of a `pa.ListArray` have the specified length."""
1811
+ return pc.all(pc.equal(array.value_lengths(), length)).as_py() or array.null_count == len(array)
1812
+
1813
+
1814
+ def _combine_list_array_offsets_with_mask(array: pa.ListArray) -> pa.Array:
1815
+ """Add the null bitmap to the offsets of a `pa.ListArray`."""
1816
+ offsets = array.offsets
1817
+ if array.null_count > 0:
1818
+ offsets = pa.concat_arrays(
1819
+ [
1820
+ pc.replace_with_mask(offsets[:-1], array.is_null(), pa.nulls(len(array), pa.int32())),
1821
+ offsets[-1:],
1822
+ ]
1823
+ )
1824
+ return offsets
1825
+
1826
+
1827
+ def _storage_type(type: pa.DataType) -> pa.DataType:
1828
+ """Convert a (possibly nested) `pa.ExtensionType` to its storage type."""
1829
+ if isinstance(type, pa.ExtensionType):
1830
+ return _storage_type(type.storage_type)
1831
+ elif isinstance(type, pa.StructType):
1832
+ return pa.struct([pa.field(field.name, _storage_type(field.type)) for field in type])
1833
+ elif isinstance(type, pa.ListType):
1834
+ return pa.list_(_storage_type(type.value_type))
1835
+ elif isinstance(type, pa.FixedSizeListType):
1836
+ return pa.list_(_storage_type(type.value_type), type.list_size)
1837
+ return type
1838
+
1839
+
1840
+ @_wrap_for_chunked_arrays
1841
+ def array_cast(
1842
+ array: pa.Array, pa_type: pa.DataType, allow_primitive_to_str: bool = True, allow_decimal_to_str: bool = True
1843
+ ) -> Union[pa.Array, pa.FixedSizeListArray, pa.ListArray, pa.StructArray, pa.ExtensionArray]:
1844
+ """Improved version of `pa.Array.cast`
1845
+
1846
+ It supports casting `pa.StructArray` objects to re-order the fields.
1847
+ It also let you control certain aspects of the casting, e.g. whether
1848
+ to disable casting primitives (`booleans`, `floats` or `ints`) or
1849
+ disable casting decimals to strings.
1850
+
1851
+ Args:
1852
+ array (`pa.Array`):
1853
+ PyArrow array to cast
1854
+ pa_type (`pa.DataType`):
1855
+ Target PyArrow type
1856
+ allow_primitive_to_str (`bool`, defaults to `True`):
1857
+ Whether to allow casting primitives to strings.
1858
+ Defaults to `True`.
1859
+ allow_decimal_to_str (`bool`, defaults to `True`):
1860
+ Whether to allow casting decimals to strings.
1861
+ Defaults to `True`.
1862
+
1863
+ Raises:
1864
+ `pa.ArrowInvalidError`: if the arrow data casting fails
1865
+ `TypeError`: if the target type is not supported according, e.g.
1866
+
1867
+ - if a field is missing
1868
+ - if casting from primitives to strings and `allow_primitive_to_str` is `False`
1869
+ - if casting from decimals to strings and `allow_decimal_to_str` is `False`
1870
+
1871
+ Returns:
1872
+ `List[pyarrow.Array]`: the casted array
1873
+ """
1874
+ _c = partial(array_cast, allow_primitive_to_str=allow_primitive_to_str, allow_decimal_to_str=allow_decimal_to_str)
1875
+ if isinstance(array, pa.ExtensionArray):
1876
+ array = array.storage
1877
+ if isinstance(pa_type, pa.ExtensionType):
1878
+ return pa_type.wrap_array(_c(array, pa_type.storage_type))
1879
+ elif array.type == pa_type:
1880
+ return array
1881
+ elif pa.types.is_struct(array.type):
1882
+ if pa.types.is_struct(pa_type) and ({field.name for field in pa_type} == {field.name for field in array.type}):
1883
+ if array.type.num_fields == 0:
1884
+ return array
1885
+ arrays = [_c(array.field(field.name), field.type) for field in pa_type]
1886
+ return pa.StructArray.from_arrays(arrays, fields=list(pa_type), mask=array.is_null())
1887
+ elif pa.types.is_list(array.type):
1888
+ if pa.types.is_fixed_size_list(pa_type):
1889
+ if _are_list_values_of_length(array, pa_type.list_size):
1890
+ if array.null_count > 0:
1891
+ # Ensure each null value in the array translates to [null] * pa_type.list_size in the array's values array
1892
+ array_type = array.type
1893
+ storage_type = _storage_type(array_type)
1894
+ if array_type != storage_type:
1895
+ # Temporarily convert to the storage type to support extension types in the slice operation
1896
+ array = _c(array, storage_type)
1897
+ array = pc.list_slice(array, 0, pa_type.list_size, return_fixed_size_list=True)
1898
+ array = _c(array, array_type)
1899
+ else:
1900
+ array = pc.list_slice(array, 0, pa_type.list_size, return_fixed_size_list=True)
1901
+ array_values = array.values
1902
+ if config.PYARROW_VERSION.major < 15:
1903
+ return pa.Array.from_buffers(
1904
+ pa_type,
1905
+ len(array),
1906
+ [array.is_valid().buffers()[1]],
1907
+ children=[_c(array_values, pa_type.value_type)],
1908
+ )
1909
+ else:
1910
+ return pa.FixedSizeListArray.from_arrays(
1911
+ _c(array_values, pa_type.value_type), pa_type.list_size, mask=array.is_null()
1912
+ )
1913
+ else:
1914
+ array_values = array.values[
1915
+ array.offset * pa_type.length : (array.offset + len(array)) * pa_type.length
1916
+ ]
1917
+ return pa.FixedSizeListArray.from_arrays(_c(array_values, pa_type.value_type), pa_type.list_size)
1918
+ elif pa.types.is_list(pa_type):
1919
+ # Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
1920
+ array_offsets = _combine_list_array_offsets_with_mask(array)
1921
+ return pa.ListArray.from_arrays(array_offsets, _c(array.values, pa_type.value_type))
1922
+ elif pa.types.is_fixed_size_list(array.type):
1923
+ if pa.types.is_fixed_size_list(pa_type):
1924
+ if pa_type.list_size == array.type.list_size:
1925
+ array_values = array.values[
1926
+ array.offset * array.type.list_size : (array.offset + len(array)) * array.type.list_size
1927
+ ]
1928
+ if config.PYARROW_VERSION.major < 15:
1929
+ return pa.Array.from_buffers(
1930
+ pa_type,
1931
+ len(array),
1932
+ [array.is_valid().buffers()[1]],
1933
+ children=[_c(array_values, pa_type.value_type)],
1934
+ )
1935
+ else:
1936
+ return pa.FixedSizeListArray.from_arrays(
1937
+ _c(array_values, pa_type.value_type), pa_type.list_size, mask=array.is_null()
1938
+ )
1939
+ elif pa.types.is_list(pa_type):
1940
+ array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
1941
+ return pa.ListArray.from_arrays(array_offsets, _c(array.values, pa_type.value_type), mask=array.is_null())
1942
+ else:
1943
+ if pa.types.is_string(pa_type):
1944
+ if not allow_primitive_to_str and pa.types.is_primitive(array.type):
1945
+ raise TypeError(
1946
+ f"Couldn't cast array of type {array.type} to {pa_type} "
1947
+ f"since allow_primitive_to_str is set to {allow_primitive_to_str} "
1948
+ )
1949
+ if not allow_decimal_to_str and pa.types.is_decimal(array.type):
1950
+ raise TypeError(
1951
+ f"Couldn't cast array of type {array.type} to {pa_type} "
1952
+ f"and allow_decimal_to_str is set to {allow_decimal_to_str}"
1953
+ )
1954
+ if pa.types.is_null(pa_type) and not pa.types.is_null(array.type):
1955
+ raise TypeError(f"Couldn't cast array of type {array.type} to {pa_type}")
1956
+ return array.cast(pa_type)
1957
+ raise TypeError(f"Couldn't cast array of type\n{array.type}\nto\n{pa_type}")
1958
+
1959
+
1960
+ @_wrap_for_chunked_arrays
1961
+ def cast_array_to_feature(
1962
+ array: pa.Array, feature: "FeatureType", allow_primitive_to_str: bool = True, allow_decimal_to_str: bool = True
1963
+ ) -> pa.Array:
1964
+ """Cast an array to the arrow type that corresponds to the requested feature type.
1965
+ For custom features like [`Audio`] or [`Image`], it takes into account the "cast_storage" methods
1966
+ they defined to enable casting from other arrow types.
1967
+
1968
+ Args:
1969
+ array (`pa.Array`):
1970
+ The PyArrow array to cast.
1971
+ feature (`datasets.features.FeatureType`):
1972
+ The target feature type.
1973
+ allow_primitive_to_str (`bool`, defaults to `True`):
1974
+ Whether to allow casting primitives to strings.
1975
+ Defaults to `True`.
1976
+ allow_decimal_to_str (`bool`, defaults to `True`):
1977
+ Whether to allow casting decimals to strings.
1978
+ Defaults to `True`.
1979
+
1980
+ Raises:
1981
+ `pa.ArrowInvalidError`: if the arrow data casting fails
1982
+ `TypeError`: if the target type is not supported according, e.g.
1983
+
1984
+ - if a field is missing
1985
+ - if casting from primitives and `allow_primitive_to_str` is `False`
1986
+ - if casting from decimals and `allow_decimal_to_str` is `False`
1987
+
1988
+ Returns:
1989
+ array (`pyarrow.Array`): the casted array
1990
+ """
1991
+ from .features.features import Sequence, get_nested_type
1992
+
1993
+ _c = partial(
1994
+ cast_array_to_feature,
1995
+ allow_primitive_to_str=allow_primitive_to_str,
1996
+ allow_decimal_to_str=allow_decimal_to_str,
1997
+ )
1998
+
1999
+ if isinstance(array, pa.ExtensionArray):
2000
+ array = array.storage
2001
+ if hasattr(feature, "cast_storage"):
2002
+ return feature.cast_storage(array)
2003
+
2004
+ elif pa.types.is_struct(array.type):
2005
+ # feature must be a dict or Sequence(subfeatures_dict)
2006
+ if isinstance(feature, Sequence) and isinstance(feature.feature, dict):
2007
+ feature = {
2008
+ name: Sequence(subfeature, length=feature.length) for name, subfeature in feature.feature.items()
2009
+ }
2010
+ if isinstance(feature, dict) and {field.name for field in array.type} == set(feature):
2011
+ if array.type.num_fields == 0:
2012
+ return array
2013
+ arrays = [_c(array.field(name), subfeature) for name, subfeature in feature.items()]
2014
+ return pa.StructArray.from_arrays(arrays, names=list(feature), mask=array.is_null())
2015
+ elif pa.types.is_list(array.type):
2016
+ # feature must be either [subfeature] or Sequence(subfeature)
2017
+ if isinstance(feature, list):
2018
+ casted_array_values = _c(array.values, feature[0])
2019
+ if casted_array_values.type == array.values.type:
2020
+ return array
2021
+ else:
2022
+ # Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
2023
+ array_offsets = _combine_list_array_offsets_with_mask(array)
2024
+ return pa.ListArray.from_arrays(array_offsets, casted_array_values)
2025
+ elif isinstance(feature, Sequence):
2026
+ if feature.length > -1:
2027
+ if _are_list_values_of_length(array, feature.length):
2028
+ if array.null_count > 0:
2029
+ # Ensure each null value in the array translates to [null] * pa_type.list_size in the array's values array
2030
+ array_type = array.type
2031
+ storage_type = _storage_type(array_type)
2032
+ if array_type != storage_type:
2033
+ # Temporarily convert to the storage type to support extension types in the slice operation
2034
+ array = array_cast(
2035
+ array,
2036
+ storage_type,
2037
+ allow_primitive_to_str=allow_primitive_to_str,
2038
+ allow_decimal_to_str=allow_decimal_to_str,
2039
+ )
2040
+ array = pc.list_slice(array, 0, feature.length, return_fixed_size_list=True)
2041
+ array = array_cast(
2042
+ array,
2043
+ array_type,
2044
+ allow_primitive_to_str=allow_primitive_to_str,
2045
+ allow_decimal_to_str=allow_decimal_to_str,
2046
+ )
2047
+ else:
2048
+ array = pc.list_slice(array, 0, feature.length, return_fixed_size_list=True)
2049
+ array_values = array.values
2050
+ casted_array_values = _c(array_values, feature.feature)
2051
+ if config.PYARROW_VERSION.major < 15:
2052
+ return pa.Array.from_buffers(
2053
+ pa.list_(casted_array_values.type, feature.length),
2054
+ len(array),
2055
+ [array.is_valid().buffers()[1]],
2056
+ children=[casted_array_values],
2057
+ )
2058
+ else:
2059
+ return pa.FixedSizeListArray.from_arrays(
2060
+ casted_array_values, feature.length, mask=array.is_null()
2061
+ )
2062
+ else:
2063
+ array_values = array.values[
2064
+ array.offset * feature.length : (array.offset + len(array)) * feature.length
2065
+ ]
2066
+ return pa.FixedSizeListArray.from_arrays(_c(array_values, feature.feature), feature.length)
2067
+ else:
2068
+ casted_array_values = _c(array.values, feature.feature)
2069
+ if casted_array_values.type == array.values.type:
2070
+ return array
2071
+ else:
2072
+ # Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
2073
+ array_offsets = _combine_list_array_offsets_with_mask(array)
2074
+ return pa.ListArray.from_arrays(array_offsets, casted_array_values)
2075
+ elif pa.types.is_fixed_size_list(array.type):
2076
+ # feature must be either [subfeature] or Sequence(subfeature)
2077
+ if isinstance(feature, list):
2078
+ array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
2079
+ return pa.ListArray.from_arrays(array_offsets, _c(array.values, feature[0]), mask=array.is_null())
2080
+ elif isinstance(feature, Sequence):
2081
+ if feature.length > -1:
2082
+ if feature.length == array.type.list_size:
2083
+ array_values = array.values[
2084
+ array.offset * array.type.list_size : (array.offset + len(array)) * array.type.list_size
2085
+ ]
2086
+ casted_array_values = _c(array_values, feature.feature)
2087
+ if config.PYARROW_VERSION.major < 15:
2088
+ return pa.Array.from_buffers(
2089
+ pa.list_(casted_array_values.type, feature.length),
2090
+ len(array),
2091
+ [array.is_valid().buffers()[1]],
2092
+ children=[casted_array_values],
2093
+ )
2094
+ else:
2095
+ return pa.FixedSizeListArray.from_arrays(
2096
+ casted_array_values, feature.length, mask=array.is_null()
2097
+ )
2098
+ else:
2099
+ array_offsets = (np.arange(len(array) + 1) + array.offset) * array.type.list_size
2100
+ return pa.ListArray.from_arrays(array_offsets, _c(array.values, feature.feature), mask=array.is_null())
2101
+ if pa.types.is_null(array.type):
2102
+ return array_cast(
2103
+ array,
2104
+ get_nested_type(feature),
2105
+ allow_primitive_to_str=allow_primitive_to_str,
2106
+ allow_decimal_to_str=allow_decimal_to_str,
2107
+ )
2108
+ elif not isinstance(feature, (Sequence, dict, list, tuple)):
2109
+ return array_cast(
2110
+ array,
2111
+ feature(),
2112
+ allow_primitive_to_str=allow_primitive_to_str,
2113
+ allow_decimal_to_str=allow_decimal_to_str,
2114
+ )
2115
+ raise TypeError(f"Couldn't cast array of type\n{array.type}\nto\n{feature}")
2116
+
2117
+
2118
+ @_wrap_for_chunked_arrays
2119
+ def embed_array_storage(array: pa.Array, feature: "FeatureType"):
2120
+ """Embed data into an arrays's storage.
2121
+ For custom features like Audio or Image, it takes into account the "embed_storage" methods
2122
+ they define to embed external data (e.g. an image file) into an array.
2123
+
2124
+ <Added version="2.4.0"/>
2125
+
2126
+ Args:
2127
+ array (`pa.Array`):
2128
+ The PyArrow array in which to embed data.
2129
+ feature (`datasets.features.FeatureType`):
2130
+ Array features.
2131
+
2132
+ Raises:
2133
+ `TypeError`: if the target type is not supported according, e.g.
2134
+
2135
+ - if a field is missing
2136
+
2137
+ Returns:
2138
+ array (`pyarrow.Array`): the casted array
2139
+ """
2140
+ from .features import Sequence
2141
+
2142
+ _e = embed_array_storage
2143
+
2144
+ if isinstance(array, pa.ExtensionArray):
2145
+ array = array.storage
2146
+ if hasattr(feature, "embed_storage"):
2147
+ return feature.embed_storage(array)
2148
+ elif pa.types.is_struct(array.type):
2149
+ # feature must be a dict or Sequence(subfeatures_dict)
2150
+ if isinstance(feature, Sequence) and isinstance(feature.feature, dict):
2151
+ feature = {
2152
+ name: Sequence(subfeature, length=feature.length) for name, subfeature in feature.feature.items()
2153
+ }
2154
+ if isinstance(feature, dict):
2155
+ arrays = [_e(array.field(name), subfeature) for name, subfeature in feature.items()]
2156
+ return pa.StructArray.from_arrays(arrays, names=list(feature), mask=array.is_null())
2157
+ elif pa.types.is_list(array.type):
2158
+ # feature must be either [subfeature] or Sequence(subfeature)
2159
+ # Merge offsets with the null bitmap to avoid the "Null bitmap with offsets slice not supported" ArrowNotImplementedError
2160
+ array_offsets = _combine_list_array_offsets_with_mask(array)
2161
+ if isinstance(feature, list):
2162
+ return pa.ListArray.from_arrays(array_offsets, _e(array.values, feature[0]))
2163
+ if isinstance(feature, Sequence) and feature.length == -1:
2164
+ return pa.ListArray.from_arrays(array_offsets, _e(array.values, feature.feature))
2165
+ elif pa.types.is_fixed_size_list(array.type):
2166
+ # feature must be Sequence(subfeature)
2167
+ if isinstance(feature, Sequence) and feature.length > -1:
2168
+ array_values = array.values[
2169
+ array.offset * array.type.list_size : (array.offset + len(array)) * array.type.list_size
2170
+ ]
2171
+ embedded_array_values = _e(array_values, feature.feature)
2172
+ if config.PYARROW_VERSION.major < 15:
2173
+ return pa.Array.from_buffers(
2174
+ pa.list_(array_values.type, feature.length),
2175
+ len(array),
2176
+ [array.is_valid().buffers()[1]],
2177
+ children=[embedded_array_values],
2178
+ )
2179
+ else:
2180
+ return pa.FixedSizeListArray.from_arrays(embedded_array_values, feature.length, mask=array.is_null())
2181
+ if not isinstance(feature, (Sequence, dict, list, tuple)):
2182
+ return array
2183
+ raise TypeError(f"Couldn't embed array of type\n{array.type}\nwith\n{feature}")
2184
+
2185
+
2186
+ class CastError(ValueError):
2187
+ """When it's not possible to cast an Arrow table to a specific schema or set of features"""
2188
+
2189
+ def __init__(self, *args, table_column_names: List[str], requested_column_names: List[str]) -> None:
2190
+ super().__init__(*args)
2191
+ self.table_column_names = table_column_names
2192
+ self.requested_column_names = requested_column_names
2193
+
2194
+ def __reduce__(self):
2195
+ # Fix unpickling: TypeError: __init__() missing 2 required keyword-only arguments: 'table_column_names' and 'requested_column_names'
2196
+ return partial(
2197
+ CastError, table_column_names=self.table_column_names, requested_column_names=self.requested_column_names
2198
+ ), ()
2199
+
2200
+ def details(self):
2201
+ new_columns = set(self.table_column_names) - set(self.requested_column_names)
2202
+ missing_columns = set(self.requested_column_names) - set(self.table_column_names)
2203
+ if new_columns and missing_columns:
2204
+ return f"there are {len(new_columns)} new columns ({', '.join(new_columns)}) and {len(missing_columns)} missing columns ({', '.join(missing_columns)})."
2205
+ elif new_columns:
2206
+ return f"there are {len(new_columns)} new columns ({new_columns})"
2207
+ else:
2208
+ return f"there are {len(missing_columns)} missing columns ({missing_columns})"
2209
+
2210
+
2211
+ def cast_table_to_features(table: pa.Table, features: "Features"):
2212
+ """Cast a table to the arrow schema that corresponds to the requested features.
2213
+
2214
+ Args:
2215
+ table (`pyarrow.Table`):
2216
+ PyArrow table to cast.
2217
+ features ([`Features`]):
2218
+ Target features.
2219
+
2220
+ Returns:
2221
+ table (`pyarrow.Table`): the casted table
2222
+ """
2223
+ if sorted(table.column_names) != sorted(features):
2224
+ raise CastError(
2225
+ f"Couldn't cast\n{table.schema}\nto\n{features}\nbecause column names don't match",
2226
+ table_column_names=table.column_names,
2227
+ requested_column_names=list(features),
2228
+ )
2229
+ arrays = [cast_array_to_feature(table[name], feature) for name, feature in features.items()]
2230
+ return pa.Table.from_arrays(arrays, schema=features.arrow_schema)
2231
+
2232
+
2233
+ def cast_table_to_schema(table: pa.Table, schema: pa.Schema):
2234
+ """Cast a table to the arrow schema. Different from `cast_table_to_features`, this method can preserve nullability.
2235
+
2236
+ Args:
2237
+ table (`pa.Table`):
2238
+ PyArrow table to cast.
2239
+ features ([`Features`]):
2240
+ Target features.
2241
+
2242
+ Returns:
2243
+ `pa.Table`: the casted table
2244
+ """
2245
+ from .features import Features
2246
+
2247
+ features = Features.from_arrow_schema(schema)
2248
+ if sorted(table.column_names) != sorted(features):
2249
+ raise CastError(
2250
+ f"Couldn't cast\n{table.schema}\nto\n{features}\nbecause column names don't match",
2251
+ table_column_names=table.column_names,
2252
+ requested_column_names=list(features),
2253
+ )
2254
+ arrays = [cast_array_to_feature(table[name], feature) for name, feature in features.items()]
2255
+ return pa.Table.from_arrays(arrays, schema=schema)
2256
+
2257
+
2258
+ def embed_table_storage(table: pa.Table):
2259
+ """Embed external data into a table's storage.
2260
+
2261
+ <Added version="2.4.0"/>
2262
+
2263
+ Args:
2264
+ table (`pyarrow.Table`):
2265
+ PyArrow table in which to embed data.
2266
+
2267
+ Returns:
2268
+ table (`pyarrow.Table`): the table with embedded data
2269
+ """
2270
+ from .features.features import Features, require_storage_embed
2271
+
2272
+ features = Features.from_arrow_schema(table.schema)
2273
+ arrays = [
2274
+ embed_array_storage(table[name], feature) if require_storage_embed(feature) else table[name]
2275
+ for name, feature in features.items()
2276
+ ]
2277
+ return pa.Table.from_arrays(arrays, schema=features.arrow_schema)
2278
+
2279
+
2280
+ def table_cast(table: pa.Table, schema: pa.Schema):
2281
+ """Improved version of `pa.Table.cast`.
2282
+
2283
+ It supports casting to feature types stored in the schema metadata.
2284
+
2285
+ Args:
2286
+ table (`pyarrow.Table`):
2287
+ PyArrow table to cast.
2288
+ schema (`pyarrow.Schema`):
2289
+ Target PyArrow schema.
2290
+
2291
+ Returns:
2292
+ table (`pyarrow.Table`): the casted table
2293
+ """
2294
+ if table.schema != schema:
2295
+ return cast_table_to_schema(table, schema)
2296
+ elif table.schema.metadata != schema.metadata:
2297
+ return table.replace_schema_metadata(schema.metadata)
2298
+ else:
2299
+ return table
2300
+
2301
+
2302
+ def table_flatten(table: pa.Table):
2303
+ """Improved version of `pa.Table.flatten`.
2304
+
2305
+ It behaves as `pa.Table.flatten` in a sense it does 1-step flatten of the columns with a struct type into one column per struct field,
2306
+ but updates the metadata and skips decodable features unless the `decode` attribute of these features is set to False.
2307
+
2308
+ Args:
2309
+ table (`pa.Table`):
2310
+ PyArrow table to flatten.
2311
+
2312
+ Returns:
2313
+ `Table`: the flattened table
2314
+ """
2315
+ from .features import Features
2316
+
2317
+ features = Features.from_arrow_schema(table.schema)
2318
+ if any(hasattr(subfeature, "flatten") and subfeature.flatten() == subfeature for subfeature in features.values()):
2319
+ flat_arrays = []
2320
+ flat_column_names = []
2321
+ for field in table.schema:
2322
+ array = table.column(field.name)
2323
+ subfeature = features[field.name]
2324
+ if pa.types.is_struct(field.type) and (
2325
+ not hasattr(subfeature, "flatten") or subfeature.flatten() != subfeature
2326
+ ):
2327
+ flat_arrays.extend(array.flatten())
2328
+ flat_column_names.extend([f"{field.name}.{subfield.name}" for subfield in field.type])
2329
+ else:
2330
+ flat_arrays.append(array)
2331
+ flat_column_names.append(field.name)
2332
+ flat_table = pa.Table.from_arrays(
2333
+ flat_arrays,
2334
+ names=flat_column_names,
2335
+ )
2336
+ else:
2337
+ flat_table = table.flatten()
2338
+ # Preserve complex types in the metadata
2339
+ flat_features = features.flatten(max_depth=2)
2340
+ flat_features = Features({column_name: flat_features[column_name] for column_name in flat_table.column_names})
2341
+ return flat_table.replace_schema_metadata(flat_features.arrow_schema.metadata)
2342
+
2343
+
2344
+ def table_visitor(table: pa.Table, function: Callable[[pa.Array], None]):
2345
+ """Visit all arrays in a table and apply a function to them.
2346
+
2347
+ Args:
2348
+ table (`pyarrow.Table`):
2349
+ PyArrow table to visit.
2350
+ function (`Callable[[pa.Array], None]`):
2351
+ Function to apply to each array.
2352
+ """
2353
+ from .features import Features, Sequence
2354
+
2355
+ features = Features.from_arrow_schema(table.schema)
2356
+
2357
+ def _visit(array, feature):
2358
+ if isinstance(array, pa.ChunkedArray):
2359
+ for chunk in array.chunks:
2360
+ _visit(chunk, feature)
2361
+ else:
2362
+ if isinstance(array, pa.ExtensionArray):
2363
+ array = array.storage
2364
+ function(array, feature)
2365
+ if pa.types.is_struct(array.type) and not hasattr(feature, "cast_storage"):
2366
+ if isinstance(feature, Sequence) and isinstance(feature.feature, dict):
2367
+ feature = {
2368
+ name: Sequence(subfeature, length=feature.length)
2369
+ for name, subfeature in feature.feature.items()
2370
+ }
2371
+ for name, subfeature in feature.items():
2372
+ _visit(array.field(name), subfeature)
2373
+ elif pa.types.is_list(array.type):
2374
+ if isinstance(feature, list):
2375
+ _visit(array.values, feature[0])
2376
+ elif isinstance(feature, Sequence):
2377
+ _visit(array.values, feature.feature)
2378
+
2379
+ for name, feature in features.items():
2380
+ _visit(table[name], feature)
2381
+
2382
+
2383
+ def table_iter(table: Table, batch_size: int, drop_last_batch=False) -> Iterator[pa.Table]:
2384
+ """Iterate over sub-tables of size `batch_size`.
2385
+
2386
+ Args:
2387
+ table (`pyarrow.Table`):
2388
+ PyArrow table to iterate over.
2389
+ batch_size (`int`):
2390
+ Size of each sub-table to yield.
2391
+ drop_last_batch (`bool`, defaults to `False`):
2392
+ Drop the last batch if it is smaller than `batch_size`.
2393
+ """
2394
+ chunks_buffer = []
2395
+ chunks_buffer_size = 0
2396
+ for chunk in table.to_reader(max_chunksize=batch_size):
2397
+ if len(chunk) == 0:
2398
+ continue
2399
+ elif chunks_buffer_size + len(chunk) < batch_size:
2400
+ chunks_buffer.append(chunk)
2401
+ chunks_buffer_size += len(chunk)
2402
+ continue
2403
+ elif chunks_buffer_size + len(chunk) == batch_size:
2404
+ chunks_buffer.append(chunk)
2405
+ yield pa.Table.from_batches(chunks_buffer)
2406
+ chunks_buffer = []
2407
+ chunks_buffer_size = 0
2408
+ else:
2409
+ cropped_chunk_length = batch_size - chunks_buffer_size
2410
+ chunks_buffer.append(chunk.slice(0, cropped_chunk_length))
2411
+ yield pa.Table.from_batches(chunks_buffer)
2412
+ chunks_buffer = [chunk.slice(cropped_chunk_length, len(chunk) - cropped_chunk_length)]
2413
+ chunks_buffer_size = len(chunk) - cropped_chunk_length
2414
+ if not drop_last_batch and chunks_buffer:
2415
+ yield pa.Table.from_batches(chunks_buffer)
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Bermuda ADDED
Binary file (2.4 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Canary ADDED
Binary file (1.9 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Cape_Verde ADDED
Binary file (256 Bytes). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Faeroe ADDED
Binary file (1.82 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Faroe ADDED
Binary file (1.82 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Jan_Mayen ADDED
Binary file (2.3 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Reykjavik ADDED
Binary file (148 Bytes). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/South_Georgia ADDED
Binary file (150 Bytes). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Atlantic/Stanley ADDED
Binary file (1.2 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Mexico/BajaNorte ADDED
Binary file (2.37 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Mexico/BajaSur ADDED
Binary file (1.13 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Mexico/General ADDED
Binary file (1.22 kB). View file
 
venv/lib/python3.10/site-packages/pytz/zoneinfo/Pacific/Apia ADDED
Binary file (598 Bytes). View file