Spaces:
Running
Running
File size: 8,977 Bytes
717901c c46b826 | 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 | # text_engine.py
import os
import pickle
import logging
from typing import List, Optional
import numpy as np
from sentence_transformers import SentenceTransformer
import faiss
from rank_bm25 import BM25Okapi
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class Text_Search_Engine:
def __init__(
self,
base_folder: str = "vector_store",
model_name: str = "sentence-transformers/LaBSE",
index_type: str = "flat",
):
self.base_folder = base_folder
self.embeddings_folder = os.path.join(base_folder, "embeddings")
self.docs_folder = os.path.join(base_folder, "documents")
os.makedirs(self.embeddings_folder, exist_ok=True)
os.makedirs(self.docs_folder, exist_ok=True)
self.model = SentenceTransformer(model_name)
self.index: Optional[faiss.Index] = None
self.rows: List[dict] = []
self.texts: List[str] = []
self.bm25: Optional[BM25Okapi] = None
self.index_type = index_type
# -------------------------
# Index creation utilities
# -------------------------
def _create_index(self, dimension: int, embeddings: np.ndarray):
if self.index_type == "flat":
self.index = faiss.IndexFlatL2(dimension)
elif self.index_type == "ivf":
nlist = max(1, min(256, len(embeddings) // 10))
quantizer = faiss.IndexFlatL2(dimension)
self.index = faiss.IndexIVFFlat(quantizer, dimension, nlist, faiss.METRIC_L2)
self.index.train(np.array(embeddings).astype("float32"))
elif self.index_type == "hnsw":
self.index = faiss.IndexHNSWFlat(dimension, 32)
else:
raise ValueError(f"Unsupported index type: {self.index_type}")
def _persist(self):
try:
if self.index is not None:
faiss.write_index(self.index, os.path.join(self.embeddings_folder, "multilingual.index"))
with open(os.path.join(self.docs_folder, "rows.pkl"), "wb") as f:
pickle.dump(self.rows, f)
logger.info("Persisted index and rows to disk.")
except Exception as e:
logger.exception("Failed to persist index/rows: %s", e)
# -------------------------
# Core operations
# -------------------------
def encode_store(self, rows: List[dict], texts: List[str]):
try:
embeddings = self.model.encode(texts, convert_to_numpy=True)
dimension = embeddings.shape[1]
self._create_index(dimension, embeddings)
self.index.add(np.array(embeddings).astype("float32"))
self.rows = rows
self.texts = texts
tokenized_corpus = [t.lower().split() for t in texts]
self.bm25 = BM25Okapi(tokenized_corpus)
self._persist()
logger.info("Index built with %d rows (index_type=%s).", len(rows), self.index_type)
except Exception as e:
logger.exception("Error in encode_store: %s", e)
raise
def load(self):
try:
index_path = os.path.join(self.embeddings_folder, "multilingual.index")
rows_path = os.path.join(self.docs_folder, "rows.pkl")
if os.path.exists(index_path) and os.path.exists(rows_path):
self.index = faiss.read_index(index_path)
with open(rows_path, "rb") as f:
self.rows = pickle.load(f)
self.texts = [r["_search_text"] for r in self.rows]
tokenized_corpus = [t.lower().split() for t in self.texts]
self.bm25 = BM25Okapi(tokenized_corpus)
logger.info("Loaded index and %d rows from disk.", len(self.rows))
else:
logger.info("No persisted index/rows found.")
except Exception as e:
logger.exception("Error in load: %s", e)
raise
def add_rows(self, new_rows: List[dict], new_texts: List[str]):
try:
if not new_rows:
return
new_embeddings = self.model.encode(new_texts, convert_to_numpy=True).astype("float32")
if self.index is None:
self._create_index(new_embeddings.shape[1], new_embeddings)
self.index.add(new_embeddings)
else:
if isinstance(self.index, faiss.IndexIVFFlat) and not self.index.is_trained:
combined = np.vstack([self.model.encode(self.texts, convert_to_numpy=True).astype("float32"), new_embeddings]) if self.texts else new_embeddings
self.index.train(combined)
self.index.add(new_embeddings)
self.rows.extend(new_rows)
self.texts.extend(new_texts)
tokenized_corpus = [t.lower().split() for t in self.texts]
self.bm25 = BM25Okapi(tokenized_corpus)
self._persist()
logger.info("Added %d new rows. Total rows: %d", len(new_rows), len(self.rows))
except Exception as e:
logger.exception("Error in add_rows: %s", e)
raise
# -------------------------
# Search methods
# -------------------------
def search(self, query: str, top_k: int = 3):
try:
if self.index is None:
return []
query_emb = self.model.encode([query], convert_to_numpy=True).astype("float32")
k = min(top_k, len(self.rows))
distances, indices = self.index.search(query_emb, k=k)
results = [
{**self.rows[i], "distance": float(distances[0][j])}
for j, i in enumerate(indices[0])
]
return sorted(results, key=lambda x: x["distance"])
except Exception as e:
logger.exception("Error in search: %s", e)
return []
def hybrid_search(self, query: str, top_k: int = 3, alpha: float = 0.5):
try:
if self.index is None or self.bm25 is None:
return []
# 🔹 Step 1: Encode query
query_emb = self.model.encode([query], convert_to_numpy=True).astype("float32")
# 🔹 Step 2: Retrieve top candidates (IMPORTANT)
retrieve_k = min(20, len(self.texts)) # candidate pool
distances, indices = self.index.search(query_emb, k=retrieve_k)
candidate_ids = indices[0]
# 🔹 Step 3: Semantic scores (convert distance → similarity)
sem_scores = {}
for j, i in enumerate(candidate_ids):
sim = 1 / (1 + distances[0][j])
sem_scores[i] = sim
# 🔹 Step 4: BM25 scores (only for candidates)
tokenized_query = query.lower().split()
bm25_scores = self.bm25.get_scores(tokenized_query)
lex_scores = {i: bm25_scores[i] for i in candidate_ids}
# 🔹 Step 5: NORMALIZATION (CRITICAL)
def normalize(scores_dict):
vals = list(scores_dict.values())
if not vals:
return scores_dict
min_v, max_v = min(vals), max(vals)
if max_v - min_v == 0:
return {k: 0.0 for k in scores_dict}
return {k: (v - min_v) / (max_v - min_v) for k, v in scores_dict.items()}
sem_scores = normalize(sem_scores)
lex_scores = normalize(lex_scores)
# 🔹 Step 6: Combine scores
combined = []
for i in candidate_ids:
sem = sem_scores.get(i, 0.0)
lex = lex_scores.get(i, 0.0)
score = alpha * sem + (1 - alpha) * lex
combined.append({**self.rows[i], "score": float(score)})
# 🔹 Step 7: Sort and return
combined = sorted(combined, key=lambda x: x["score"], reverse=True)
return combined[:top_k]
except Exception as e:
logger.exception("Error in hybrid_search: %s", e)
return []
# -------------------------
# Utilities
# -------------------------
def clear_vdb(self):
try:
if self.index is not None:
try:
self.index.reset()
except Exception:
self.index = None
self.rows = []
self.texts = []
self.bm25 = None
index_path = os.path.join(self.embeddings_folder, "multilingual.index")
docs_path = os.path.join(self.docs_folder, "rows.pkl")
if os.path.exists(index_path):
os.remove(index_path)
if os.path.exists(docs_path):
os.remove(docs_path)
logger.info("Cleared vector DB and persisted files.")
except Exception as e:
logger.exception("Error in clear_vdb: %s", e)
raise |