File size: 6,209 Bytes
014f83b cf7c6f4 014f83b cf7c6f4 014f83b | 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 | """
Phase 2: RAG Engine — FAISS Vector Database for Medical Knowledge Retrieval.
Builds a FAISS index from doctor responses in the training set.
At inference, retrieves Top-K relevant medical passages for a patient query.
Usage:
# Build index
python rag_engine.py --mode build --csv ./data/train.csv
# Query index
python rag_engine.py --mode query --query "I have chest pain and shortness of breath"
"""
import argparse
import os
import numpy as np
import pandas as pd
import faiss
from sentence_transformers import SentenceTransformer
from tqdm import tqdm
import sys
sys.path.insert(0, os.path.dirname(__file__))
from config import (
SENTENCE_EMBED_MODEL,
FAISS_INDEX_PATH,
RAG_TOP_K,
RAG_EMBED_DIM,
)
class RAGEngine:
"""FAISS-backed retrieval engine for medical passages."""
def __init__(self, index_path=FAISS_INDEX_PATH, passages_path=None):
self.index_path = index_path
# If the path is passed directly (like in Hugging Face), use it!
if passages_path:
self.passages_path = passages_path
else:
self.passages_path = index_path.replace(".bin", "_passages.npy")
self.embedder = SentenceTransformer(SENTENCE_EMBED_MODEL)
self.index = None
self.passages = None
# ----------------------------------------------------------
# Build
# ----------------------------------------------------------
def build_index(self, passages: list[str], batch_size: int = 256):
"""
Encode all passages and build a FAISS Inner-Product index.
"""
print(f"Encoding {len(passages)} passages...")
embeddings = self.embedder.encode(
passages,
batch_size=batch_size,
show_progress_bar=True,
normalize_embeddings=True,
)
embeddings = np.array(embeddings, dtype="float32")
# Build FAISS index (Inner Product = cosine similarity on normalized vecs)
self.index = faiss.IndexFlatIP(RAG_EMBED_DIM)
self.index.add(embeddings)
self.passages = np.array(passages)
# Save
os.makedirs(os.path.dirname(self.index_path) or ".", exist_ok=True)
faiss.write_index(self.index, self.index_path)
np.save(self.passages_path, self.passages)
print(f"✅ FAISS index built: {self.index.ntotal} vectors")
print(f" Saved to: {self.index_path}")
# ----------------------------------------------------------
# Load
# ----------------------------------------------------------
def load_index(self):
"""Load a previously built FAISS index."""
if not os.path.exists(self.index_path):
raise FileNotFoundError(f"No FAISS index at {self.index_path}")
self.index = faiss.read_index(self.index_path)
self.passages = np.load(self.passages_path, allow_pickle=True)
print(f"Loaded FAISS index: {self.index.ntotal} vectors")
# ----------------------------------------------------------
# Retrieve
# ----------------------------------------------------------
def retrieve(self, query: str, top_k: int = RAG_TOP_K) -> list[dict]:
"""
Returns top-K passages with scores.
Returns:
list of {"passage": str, "score": float}
"""
if self.index is None:
self.load_index()
q_emb = self.embedder.encode(
[query], normalize_embeddings=True
).astype("float32")
scores, indices = self.index.search(q_emb, top_k)
results = []
for score, idx in zip(scores[0], indices[0]):
if idx < len(self.passages):
results.append({
"passage": str(self.passages[idx]),
"score": float(score),
})
return results
def retrieve_batch(self, queries: list[str], top_k: int = RAG_TOP_K):
"""Batch retrieval for training efficiency."""
if self.index is None:
self.load_index()
q_embs = self.embedder.encode(
queries, normalize_embeddings=True, batch_size=128
).astype("float32")
scores, indices = self.index.search(q_embs, top_k)
all_results = []
for i in range(len(queries)):
results = []
for score, idx in zip(scores[i], indices[i]):
if idx < len(self.passages):
results.append({
"passage": str(self.passages[idx]),
"score": float(score),
})
all_results.append(results)
return all_results
# ============================================================
# CLI
# ============================================================
def main(args):
engine = RAGEngine(args.index_path)
if args.mode == "build":
df = pd.read_csv(args.csv)
# Use the 'description' field (short, searchable) concatenated
# with the doctor's response for maximum retrieval quality
passages = []
for _, row in df.iterrows():
desc = str(row.get("description", ""))
resp = str(row.get("doctor_response", ""))
if len(resp) > 20:
passages.append(f"{desc} | {resp}")
engine.build_index(passages, batch_size=args.batch_size)
elif args.mode == "query":
results = engine.retrieve(args.query, top_k=args.top_k)
print(f"\nTop-{args.top_k} results for: '{args.query}'\n")
for i, r in enumerate(results):
print(f" [{i+1}] (score={r['score']:.4f})")
print(f" {r['passage'][:200]}...")
print()
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--mode", choices=["build", "query"], required=True)
parser.add_argument("--csv", default="./data/train.csv")
parser.add_argument("--query", default="")
parser.add_argument("--top_k", type=int, default=RAG_TOP_K)
parser.add_argument("--batch_size", type=int, default=256)
parser.add_argument("--index_path", default=FAISS_INDEX_PATH)
args = parser.parse_args()
main(args)
|