Spaces:
Running
Running
Update search_utils.py
Browse files- search_utils.py +109 -1
search_utils.py
CHANGED
@@ -4,6 +4,62 @@ import faiss
|
|
4 |
from pathlib import Path
|
5 |
from sentence_transformers import SentenceTransformer, util
|
6 |
import streamlit as st
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
|
8 |
class SemanticSearch:
|
9 |
def __init__(self, shard_dir="compressed_shards"):
|
@@ -11,6 +67,7 @@ class SemanticSearch:
|
|
11 |
self.shard_dir.mkdir(exist_ok=True, parents=True)
|
12 |
self.model = None
|
13 |
self.index_shards = []
|
|
|
14 |
|
15 |
@st.cache_resource
|
16 |
def load_model(_self):
|
@@ -61,4 +118,55 @@ class SemanticSearch:
|
|
61 |
"""Threshold-filtered search"""
|
62 |
results = self.search(query, top_k*2)
|
63 |
filtered = results[results['similarity'] > similarity_threshold].head(top_k)
|
64 |
-
return filtered.reset_index(drop=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
from pathlib import Path
|
5 |
from sentence_transformers import SentenceTransformer, util
|
6 |
import streamlit as st
|
7 |
+
import zipfile
|
8 |
+
import pandas as pd
|
9 |
+
from pathlib import Path
|
10 |
+
import streamlit as st
|
11 |
+
|
12 |
+
class MetadataManager:
|
13 |
+
def __init__(self):
|
14 |
+
self.shard_dir = Path("metadata_shards")
|
15 |
+
self.shard_map = {}
|
16 |
+
self.loaded_shards = {}
|
17 |
+
self._ensure_unzipped()
|
18 |
+
self._build_shard_map()
|
19 |
+
|
20 |
+
def _ensure_unzipped(self):
|
21 |
+
"""Extract metadata shards from zip if needed"""
|
22 |
+
if not self.shard_dir.exists():
|
23 |
+
zip_path = Path("metadata_shards.zip")
|
24 |
+
if zip_path.exists():
|
25 |
+
with zipfile.ZipFile(zip_path, 'r') as zip_ref:
|
26 |
+
zip_ref.extractall(self.shard_dir)
|
27 |
+
st.toast("✅ Successfully extracted metadata shards!", icon="📦")
|
28 |
+
else:
|
29 |
+
raise FileNotFoundError("No metadata shards found!")
|
30 |
+
|
31 |
+
def _build_shard_map(self):
|
32 |
+
"""Map index ranges to shard files"""
|
33 |
+
for f in self.shard_dir.glob("*.parquet"):
|
34 |
+
parts = f.stem.split("_")
|
35 |
+
self.shard_map[(int(parts[1]), int(parts[2]))] = f.name
|
36 |
+
|
37 |
+
def get_metadata(self, indices):
|
38 |
+
"""Retrieve metadata for specific indices"""
|
39 |
+
results = []
|
40 |
+
shard_groups = {}
|
41 |
+
|
42 |
+
# Group indices by shard
|
43 |
+
for idx in indices:
|
44 |
+
for (start, end), shard in self.shard_map.items():
|
45 |
+
if start <= idx <= end:
|
46 |
+
if shard not in shard_groups:
|
47 |
+
shard_groups[shard] = []
|
48 |
+
shard_groups[shard].append(idx - start)
|
49 |
+
break
|
50 |
+
|
51 |
+
# Load required shards
|
52 |
+
for shard, local_indices in shard_groups.items():
|
53 |
+
if shard not in self.loaded_shards:
|
54 |
+
self.loaded_shards[shard] = pd.read_parquet(
|
55 |
+
self.shard_dir / shard,
|
56 |
+
columns=["title", "summary", "source"]
|
57 |
+
)
|
58 |
+
|
59 |
+
results.append(self.loaded_shards[shard].iloc[local_indices])
|
60 |
+
|
61 |
+
return pd.concat(results).reset_index(drop=True)
|
62 |
+
|
63 |
|
64 |
class SemanticSearch:
|
65 |
def __init__(self, shard_dir="compressed_shards"):
|
|
|
67 |
self.shard_dir.mkdir(exist_ok=True, parents=True)
|
68 |
self.model = None
|
69 |
self.index_shards = []
|
70 |
+
self.metadata_mgr = MetadataManager()
|
71 |
|
72 |
@st.cache_resource
|
73 |
def load_model(_self):
|
|
|
118 |
"""Threshold-filtered search"""
|
119 |
results = self.search(query, top_k*2)
|
120 |
filtered = results[results['similarity'] > similarity_threshold].head(top_k)
|
121 |
+
return filtered.reset_index(drop=True)
|
122 |
+
|
123 |
+
|
124 |
+
|
125 |
+
|
126 |
+
class MetadataManager:
|
127 |
+
def __init__(self, repo_id, shard_dir="metadata_shards"):
|
128 |
+
self.repo_id = repo_id
|
129 |
+
self.shard_dir = Path(shard_dir)
|
130 |
+
self.shard_map = {}
|
131 |
+
self.loaded_shards = {}
|
132 |
+
self._build_shard_map()
|
133 |
+
|
134 |
+
def _build_shard_map(self):
|
135 |
+
"""Map index ranges to shard files"""
|
136 |
+
for f in self.shard_dir.glob("*.parquet"):
|
137 |
+
parts = f.stem.split("_")
|
138 |
+
self.shard_map[(int(parts[1]), int(parts[2]))] = f.name
|
139 |
+
|
140 |
+
def _download_shard(self, shard_name):
|
141 |
+
"""Download missing shards on demand"""
|
142 |
+
if not (self.shard_dir/shard_name).exists():
|
143 |
+
hf_hub_download(
|
144 |
+
repo_id=self.repo_id,
|
145 |
+
filename=f"metadata_shards/{shard_name}",
|
146 |
+
local_dir=self.shard_dir,
|
147 |
+
cache_dir="metadata_cache"
|
148 |
+
)
|
149 |
+
|
150 |
+
def get_metadata(self, indices):
|
151 |
+
"""Retrieve metadata for specific indices"""
|
152 |
+
results = []
|
153 |
+
|
154 |
+
# Group indices by shard
|
155 |
+
shard_groups = {}
|
156 |
+
for idx in indices:
|
157 |
+
for (start, end), shard in self.shard_map.items():
|
158 |
+
if start <= idx <= end:
|
159 |
+
if shard not in shard_groups:
|
160 |
+
shard_groups[shard] = []
|
161 |
+
shard_groups[shard].append(idx - start)
|
162 |
+
break
|
163 |
+
|
164 |
+
# Process each required shard
|
165 |
+
for shard, local_indices in shard_groups.items():
|
166 |
+
if shard not in self.loaded_shards:
|
167 |
+
self._download_shard(shard)
|
168 |
+
self.loaded_shards[shard] = pd.read_parquet(self.shard_dir/shard)
|
169 |
+
|
170 |
+
results.append(self.loaded_shards[shard].iloc[local_indices])
|
171 |
+
|
172 |
+
return pd.concat(results).reset_index(drop=True)
|