File size: 9,332 Bytes
cf5659c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 |
import csv
import json
import re
import subprocess
from argparse import ArgumentParser
import matplotlib.pyplot as plt
from pathlib import Path
import numpy as np
"""
Plot results per (dataset_name, dataset_config_name).
"""
def get_args():
parser = ArgumentParser()
parser.add_argument("--json_paths", nargs="+", type=str, help="Json files to plot together", required=True)
parser.add_argument("--t0_csv_path", type=str, help="T0 eval results path")
args = parser.parse_args()
return args
def load_t0_results(csv_path):
with open(csv_path, "r") as f:
return list(csv.DictReader(f))
def load_json(json_path):
with open(json_path, "r") as fi:
return json.load(fi)
def get_experiment_name(filename: str):
name = re.sub(r"_([0-9]*)$", r" [\1]", filename)
name = name.replace("span_corruption", "SC")
name = re.sub(r"^enc_dec", "ED", name)
name = re.sub(r"^nc_dec", "NCD", name)
name = re.sub(r"^c_dec", 'CD', name)
name = name.replace("full_lm", "FLM")
name = name.replace("prefix_lm", "PLM")
name = re.sub(r"t0_adapt_([0-9]+)", r"T0(\1)", name)
if name[:3] == "CD_":
name = re.sub(r"lm_adapt_([0-9]+)", r"FLM(\1)", name)
name = re.sub(r"t0_adapt_nc_([0-9]+)", r"T0 AS NC (\1)", name)
name = re.sub(r"nc_sc_([0-9]+)", r"SC as NC(\1)", name)
name = re.sub(r"nc_t0_([0-9]+)", r"T0 as NC(\1)", name)
elif name[:4] == "NCD_" or name[:3] == "ED_":
if "flm_adapt" in name:
name = re.sub(r"flm_adapt_([0-9]+)", r"FLM AS CD(\1)", name)
else:
name = re.sub(r"lm_adapt_([0-9]+)", r"PLM(\1)", name)
else:
raise NotImplementedError
name = name.replace("_", " + ")
return name
TASKS = {
# T0 evaluation
"super_glue_copa": ("COPA", 0.5),
"anli_dev_r1": ("ANLI R1", 1/3),
"anli_dev_r2": ("ANLI R2", 1/3),
"anli_dev_r3": ("ANLI R3", 1/3),
"super_glue_cb": ("CB", 1/3),
"super_glue_rte": ("RTE", 0.5),
"super_glue_wsc.fixed": ("WSC", 0.5),
"winogrande_winogrande_xl": ("Winogrande", 0.5),
"super_glue_wic": ("WiC", 0.5),
"hellaswag": ("HellaSwag", 0.25),
"story_cloze_2016": ("StoryCloze", 0.5),
# XNLI evaluation
"xnli_ar": ("XNLI ar (en prompts)", 1/3),
"xnli_bg": ("XNLI bg (en prompts)", 1/3),
"xnli_de": ("XNLI de (en prompts)", 1/3),
"xnli_el": ("XNLI el (en prompts)", 1/3),
"xnli_en": ("XNLI en (en prompts)", 1/3),
"xnli_es": ("XNLI es (en prompts)", 1/3),
"xnli_fr": ("XNLI fr (en prompts)", 1/3),
"xnli_hi": ("XNLI hi (en prompts)", 1/3),
"xnli_ru": ("XNLI ru (en prompts)", 1/3),
"xnli_sw": ("XNLI sw (en prompts)", 1/3),
"xnli_th": ("XNLI th (en prompts)", 1/3),
"xnli_tr": ("XNLI tr (en prompts)", 1/3),
"xnli_ur": ("XNLI ur (en prompts)", 1/3),
"xnli_vi": ("XNLI vi (en prompts)", 1/3),
"xnli_zh": ("XNLI zh (en prompts)", 1/3),
}
def plot(mtf_data, t0_data):
args = get_args()
assert len(TASKS) == 26
fig, axs = plt.subplots(3, 9, figsize=(20, 5))
axs = axs.flatten()
task_min_score = {}
task_max_score = {}
task_median_score = {}
for n, (task, (task_name, random_baseline)) in enumerate(TASKS.items()):
# Normalising names
mtf_task = task
t0_task = task
if task.startswith("anli_dev_r"):
t0_task = re.sub("dev_", "", task)
elif task == "hellaswag":
mtf_task = "hellaswag_None"
t5lm_scores = [float(r["score"]) for r in t0_data
if r["runs"] == "xxl-lm-d4-091621"
and r["dataset_name"] == t0_task
and r["metric_name"] == "accuracy (Rank)"
and r["score"]]
t0_scores = [float(r["score"]) for r in t0_data
if r["runs"] == "xxl-lm-d4-091621-512"
and r["dataset_name"] == t0_task
and r["metric_name"] == "accuracy (Rank)"
and r["score"]]
mtf_scores = [
(
name,
[100 * value["evaluation"]["accuracy"] for prompt, value in data[mtf_task].items()]
if mtf_task in data else
[]
)
for name, data in mtf_data.items()
]
all_experiment_scores_with_name = [("T5 + LM", t5lm_scores), ("T0", t0_scores), *mtf_scores]
# Plot
axs[n].axhline(100 * random_baseline, 0, len(all_experiment_scores_with_name), label="Random")
for i, (exp_name, scores) in enumerate(all_experiment_scores_with_name):
axs[n].scatter([i] * len(scores), scores, s=50, alpha=0.4, label=exp_name)
axs[n].set_title(task_name, fontsize=8)
# # Gather median values
# task_min_score[task] = [("Random", 100 * random_baseline)] + [(exp_name, np.min(scores)) for (exp_name, scores) in all_experiment_scores_with_name]
# task_max_score[task] = [("Random", 100 * random_baseline)] + [(exp_name, np.max(scores)) for (exp_name, scores) in all_experiment_scores_with_name]
# task_median_score[task] = [("Random", 100 * random_baseline)] + [(exp_name, np.median(scores)) for (exp_name, scores) in all_experiment_scores_with_name]
last_ax_id = len(TASKS) - 1
axs[last_ax_id].legend(bbox_to_anchor=(1, 1), loc="upper left")
for ax in axs[last_ax_id + 1:]:
ax.set_visible(False)
# if args.aggregated_results:
# # ====== Plot agregated values =======
# fig, axs = plt.subplots(1, 3, figsize=(20, 8))
# axs = axs.flatten()
# last_ax_id=0
# experiment_names = [elt[0] for elt in next(iter(task_median_score.values()))]
#
# def plot_scores_with_name(median_score_with_name, max_score, min_score, ax, title):
# assert len(median_score_with_name) == len(max_score) and len(median_score_with_name) == len(min_score)
# ax.axhline(
# median_score_with_name[0][1],
# 0, len(median_score_with_name) - 1,
# label=median_score_with_name[0][0]
# )
# for i, ((name, median_score), max_score, min_score) in enumerate(zip(median_score_with_name[1:], max_score[1:], min_score[1:])):
# ax.errorbar(
# i, median_score, ((median_score - min_score,), (max_score - median_score,)),
# fmt="o", elinewidth=1, label=name)
# ax.set_title(title)
#
# def get_average_normalised_score(task_scores):
# normalised_scores = []
# for scores_with_name in task_scores.values():
# random_name, random_baseline = scores_with_name[0]
# assert random_name == "Random"
# normalised_scores_per_task = [(scores - random_baseline) / (100 - random_baseline) for _, scores in
# scores_with_name]
# normalised_scores.append(normalised_scores_per_task)
# return np.mean(normalised_scores, axis=0)
#
# def get_average_score(task_scores):
# return np.mean(
# [[scores for _, scores in scores_with_name] for scores_with_name in task_scores.values()], axis=0)
#
# # Plot average task score
# average_task_median_score = get_average_score(task_median_score)
# assert len(experiment_names) == len(average_task_median_score)
# average_task_media_score_with_name = list(zip(experiment_names, average_task_median_score))
# del average_task_median_score
# plot_scores_with_name(
# median_score_with_name=average_task_media_score_with_name,
# max_score=get_average_score(task_max_score),
# min_score=get_average_score(task_min_score),
# ax=axs[last_ax_id],
# title=f"Average of task median scores"
# )
# last_ax_id += 1
#
# # Plot average of task median normalised scores `normalised_score = (score - random) / (1 - random)`
# average_task_normalised_median_score = get_average_normalised_score(task_median_score)
# assert len(experiment_names) == len(average_task_normalised_median_score)
# average_task_normalised_median_score_with_name = list(
# zip(experiment_names, average_task_normalised_median_score))
# del average_task_normalised_median_score
# plot_scores_with_name(
# median_score_with_name=average_task_normalised_median_score_with_name,
# max_score=get_average_normalised_score(task_max_score),
# min_score=get_average_normalised_score(task_min_score),
# ax=axs[last_ax_id],
# title=f"Average of task normalised median scores"
# )
# last_ax_id += 1
#
# axs[last_ax_id -1].legend(bbox_to_anchor=(1, 1), loc="upper left")
# for ax in axs[last_ax_id:]:
# ax.set_visible(False)
def main():
args = get_args()
# Load results
t0_data = load_t0_results(args.t0_csv_path)
mtf_data = {
re.sub(".json", "", json_path): load_json(json_path)
for json_path in args.json_paths
}
plot(mtf_data, t0_data)
plt.show()
print("Finished")
if __name__ == "__main__":
main()
|