File size: 10,692 Bytes
00eef43
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
import os
import json
import pickle
from typing import List, Dict, Tuple, Optional
from pathlib import Path
import numpy as np
from sentence_transformers import SentenceTransformer, CrossEncoder
from rank_bm25 import BM25Okapi
import chromadb


class QueryExpander:
    
    def __init__(self, openai_client):
        self.client = openai_client
    
    def expand_query(self, query: str, num_variations: int = 3) -> List[str]:
        prompt = f"""Given this user query, generate {num_variations} alternative phrasings that capture the same intent but use different words.



Original query: {query}



Return ONLY a JSON array of alternative queries, nothing else.

Example: ["query1", "query2", "query3"]"""
        
        try:
            response = self.client.chat.completions.create(
                model="gpt-4o-mini",
                messages=[{"role": "user", "content": prompt}],
                temperature=0.7
            )
            variations = json.loads(response.choices[0].message.content)
            return [query] + variations
        except Exception as e:
            print(f"Query expansion failed: {e}")
            return [query]


class HybridRetriever:
    
    def __init__(self, embedding_model: str = "all-MiniLM-L6-v2", reranker_model: str = "cross-encoder/ms-marco-MiniLM-L-6-v2", data_dir: str = "data"):
        self.data_dir = Path(data_dir)
        self.data_dir.mkdir(exist_ok=True)
        
        print("Loading embedding model...")
        self.embedder = SentenceTransformer(embedding_model)
        
        print("Loading reranker model...")
        self.reranker = CrossEncoder(reranker_model)
        
        self.chroma_client = chromadb.PersistentClient(path=str(self.data_dir / "vector_store"))
        self.documents: List[Dict] = []
        self.bm25: Optional[BM25Okapi] = None
        self.collection = None
        
    def chunk_text(self, text: str, chunk_size: int = 500, overlap: int = 50) -> List[str]:
        words = text.split()
        chunks = []
        for i in range(0, len(words), chunk_size - overlap):
            chunk = ' '.join(words[i:i + chunk_size])
            if chunk:
                chunks.append(chunk)
        return chunks
    
    def index_documents(self, documents: Dict[str, str], chunk_size: int = 500, overlap: int = 50, collection_name: str = "knowledge_base"):
        print(f"Indexing documents with chunk_size={chunk_size}, overlap={overlap}")
        
        all_chunks = []
        for doc_id, content in documents.items():
            chunks = self.chunk_text(content, chunk_size, overlap)
            for idx, chunk in enumerate(chunks):
                all_chunks.append({
                    "id": f"{doc_id}_{idx}",
                    "text": chunk,
                    "source": doc_id,
                    "chunk_idx": idx
                })
        
        self.documents = all_chunks
        
        if not all_chunks:
            raise ValueError("No text chunks created from documents. Please check your document content.")
        
        print("Building BM25 index...")
        tokenized_docs = [doc["text"].lower().split() for doc in all_chunks]
        self.bm25 = BM25Okapi(tokenized_docs)
        
        print("Building semantic index...")
        try:
            self.chroma_client.delete_collection(collection_name)
        except:
            pass
        
        self.collection = self.chroma_client.create_collection(name=collection_name, metadata={"hnsw:space": "cosine"})
        
        batch_size = 100
        for i in range(0, len(all_chunks), batch_size):
            batch = all_chunks[i:i + batch_size]
            self.collection.add(
                documents=[doc["text"] for doc in batch],
                ids=[doc["id"] for doc in batch],
                metadatas=[{"source": doc["source"], "chunk_idx": doc["chunk_idx"]} for doc in batch]
            )
        
        print(f"Indexed {len(all_chunks)} chunks from {len(documents)} documents")
        
        with open(self.data_dir / "bm25_index.pkl", "wb") as f:
            pickle.dump((self.bm25, self.documents), f)
    
    def retrieve_bm25(self, query: str, top_k: int = 10) -> List[Tuple[Dict, float]]:
        if self.bm25 is None:
            return []
        
        tokenized_query = query.lower().split()
        scores = self.bm25.get_scores(tokenized_query)
        top_indices = np.argsort(scores)[::-1][:top_k]
        
        results = []
        for idx in top_indices:
            if scores[idx] > 0:
                results.append((self.documents[idx], float(scores[idx])))
        return results
    
    def retrieve_semantic(self, query: str, top_k: int = 10) -> List[Tuple[Dict, float]]:
        if self.collection is None:
            return []
        
        results = self.collection.query(query_texts=[query], n_results=top_k)
        
        retrieved = []
        for i, doc_id in enumerate(results["ids"][0]):
            doc = next((d for d in self.documents if d["id"] == doc_id), None)
            if doc:
                distance = results["distances"][0][i]
                similarity = 1 / (1 + distance)
                retrieved.append((doc, similarity))
        return retrieved
    
    def retrieve_hybrid(self, query: str, top_k: int = 10, bm25_weight: float = 0.5, semantic_weight: float = 0.5) -> List[Tuple[Dict, float]]:
        bm25_results = self.retrieve_bm25(query, top_k * 2)
        semantic_results = self.retrieve_semantic(query, top_k * 2)
        
        def normalize_scores(results):
            if not results:
                return {}
            scores = [score for _, score in results]
            max_score = max(scores) if scores else 1.0
            min_score = min(scores) if scores else 0.0
            range_score = max_score - min_score if max_score != min_score else 1.0
            return {doc["id"]: (score - min_score) / range_score for doc, score in results}
        
        bm25_scores = normalize_scores(bm25_results)
        semantic_scores = normalize_scores(semantic_results)
        
        all_doc_ids = set(bm25_scores.keys()) | set(semantic_scores.keys())
        combined_scores = {}
        for doc_id in all_doc_ids:
            bm25_score = bm25_scores.get(doc_id, 0.0)
            semantic_score = semantic_scores.get(doc_id, 0.0)
            combined_scores[doc_id] = bm25_weight * bm25_score + semantic_weight * semantic_score
        
        sorted_ids = sorted(combined_scores.items(), key=lambda x: x[1], reverse=True)[:top_k]
        
        results = []
        for doc_id, score in sorted_ids:
            doc = next((d for d in self.documents if d["id"] == doc_id), None)
            if doc:
                results.append((doc, score))
        return results
    
    def rerank(self, query: str, documents: List[Tuple[Dict, float]], top_k: int = 5) -> List[Tuple[Dict, float]]:
        if not documents:
            return []
        
        pairs = [[query, doc["text"]] for doc, _ in documents]
        rerank_scores = self.reranker.predict(pairs)
        reranked = [(doc, float(score)) for (doc, _), score in zip(documents, rerank_scores)]
        reranked.sort(key=lambda x: x[1], reverse=True)
        return reranked[:top_k]
    
    def retrieve(self, query: str, method: str = "hybrid_rerank", top_k: int = 5, expand_query: bool = False, query_expander: Optional['QueryExpander'] = None, **kwargs) -> List[Dict]:
        queries = [query]
        
        if expand_query and query_expander:
            queries = query_expander.expand_query(query)
            print(f"Expanded to {len(queries)} queries")
        
        all_results = {}
        for q in queries:
            if method == "bm25":
                results = self.retrieve_bm25(q, top_k * 2)
            elif method == "semantic":
                results = self.retrieve_semantic(q, top_k * 2)
            elif method in ["hybrid", "hybrid_rerank"]:
                results = self.retrieve_hybrid(q, top_k * 2, kwargs.get("bm25_weight", 0.5), kwargs.get("semantic_weight", 0.5))
            else:
                raise ValueError(f"Unknown method: {method}")
            
            for doc, score in results:
                doc_id = doc["id"]
                if doc_id not in all_results:
                    all_results[doc_id] = (doc, 0.0)
                all_results[doc_id] = (doc, all_results[doc_id][1] + score)
        
        aggregated = list(all_results.values())
        aggregated.sort(key=lambda x: x[1], reverse=True)
        
        if "rerank" in method:
            print(f"Reranking {len(aggregated)} results...")
            aggregated = self.rerank(query, aggregated[:top_k * 3], top_k)
        else:
            aggregated = aggregated[:top_k]
        
        return [{"retrieval_score": score, **doc} for doc, score in aggregated]


class RAGSystem:
    
    def __init__(self, openai_client, data_dir: str = "data"):
        self.client = openai_client
        self.retriever = HybridRetriever(data_dir=data_dir)
        self.query_expander = QueryExpander(openai_client)
        self.data_dir = Path(data_dir)
        
    def load_knowledge_base(self, documents: Dict[str, str], chunk_size: int = 500, overlap: int = 50):
        self.retriever.index_documents(documents, chunk_size, overlap)
        
    def generate_answer(self, query: str, context: List[Dict], system_prompt: str) -> str:
        context_str = "\n\n".join([f"[Source: {doc['source']}, Chunk {doc['chunk_idx']}]\n{doc['text']}" for doc in context])
        
        augmented_prompt = f"""{system_prompt}



## Retrieved Context:

{context_str}



## User Query:

{query}



Please answer the query based on the context provided above."""
        
        messages = [{"role": "user", "content": augmented_prompt}]
        response = self.client.chat.completions.create(model="gpt-4o-mini", messages=messages, temperature=0.7)
        return response.choices[0].message.content
    
    def query(self, query: str, system_prompt: str, method: str = "hybrid_rerank", top_k: int = 5, expand_query: bool = False, **kwargs) -> Dict:
        context = self.retriever.retrieve(query, method=method, top_k=top_k, expand_query=expand_query, query_expander=self.query_expander if expand_query else None, **kwargs)
        answer = self.generate_answer(query, context, system_prompt)
        return {"answer": answer, "context": context, "method": method, "query": query}