File size: 8,747 Bytes
c6907ac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from utilities.constants import (
    CHUNKING_STRATEGY_TABLE_AWARE,
    CHUNKING_STRATEGY_SECTION_BASED,
    CHUNKING_STRATEGY_SEMANTIC
)
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.docstore.document import Document
from langchain_community.vectorstores import Qdrant
from langchain_openai.embeddings import OpenAIEmbeddings
import numpy as np
import pdfplumber
import re
from sentence_transformers import SentenceTransformer
from sklearn.metrics.pairwise import cosine_similarity
import tiktoken
from utilities.debugger import dprint

def create_vector_store(app_state, model_run_state, **kwargs):
    for key, value in kwargs.items():
        if hasattr(model_run_state, key):
            setattr(model_run_state, key, value)
        else:
            print(f"Warning: {key} is not an attribute of the state object")

    # Rest of your create_vector_store logic
    dprint(app_state, f"Chunk size after update: {model_run_state.chunk_size}")
    create_chunked_documents(app_state, model_run_state)

    qdrant_vectorstore = Qdrant.from_documents(
        documents=model_run_state.combined_document_objects,
        embedding=model_run_state.embedding_model,
        location=":memory:" 
    )
    qdrant_retriever = qdrant_vectorstore.as_retriever()
    model_run_state.retriever = qdrant_retriever
    print("Vector store created")

def tiktoken_len(text):
    tokens = tiktoken.encoding_for_model("gpt-4o").encode(
        text,
    )
    return len(tokens)

def create_chunked_documents(app_state, model_run_state):
    dprint(app_state, model_run_state.chunking_strategy)
    if model_run_state.chunking_strategy == CHUNKING_STRATEGY_TABLE_AWARE:
        combined_document_objects = chunk_with_table_aware(app_state, model_run_state)
    elif model_run_state.chunking_strategy == CHUNKING_STRATEGY_SECTION_BASED:
        combined_document_objects = chunk_with_section_based(app_state, model_run_state)
    elif model_run_state.chunking_strategy == CHUNKING_STRATEGY_SEMANTIC:
        combined_document_objects = chunk_with_semantic_splitter(app_state, model_run_state)
    else:
        combined_document_objects = chunk_with_recursive_splitter(app_state, model_run_state)
    model_run_state.combined_document_objects = combined_document_objects
    dprint(app_state, "Chunking completed successfully")


def chunk_with_recursive_splitter(app_state, model_run_state):    
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=model_run_state.chunk_size,  
        chunk_overlap=model_run_state.chunk_overlap,
        length_function = tiktoken_len,
    )
    combined_document_objects = []
    dprint(app_state, "Chunking documents and creating document objects")
    for document in app_state.documents:
        dprint(app_state, f"processing documend: {document['title']}")
        text = document["single_text_document"]
        dprint(app_state, text)
        title = document["title"]
        # document_id = document["document_id"]
        chunks_document = text_splitter.split_text(text)
        dprint(app_state, len(chunks_document))

        for chunk_number, chunk in enumerate(chunks_document, start=1):
            document_objects = Document(
                page_content=chunk,
                metadata={
                    "source": title,
                    "document_id": document.get("document_id", "default_id"),
                    "chunk_number": chunk_number  # Add unique chunk number
                }
            )
            combined_document_objects.append(document_objects)
    return combined_document_objects
    
def chunk_with_table_aware(app_state, model_run_state):
    combined_document_objects = []
    dprint(app_state, "Using Table-Aware Chunking for documents.")

    for document in app_state.documents:
        title = document["title"]
        text = document["single_text_document"]

        # Check if document is a PDF and contains tables
        if document.get("is_pdf", False):
            with pdfplumber.open(document["file_path"]) as pdf:
                for page in pdf.pages:
                    tables = page.extract_tables()
                    for table in tables:
                        table_content = "\n".join([str(row) for row in table])
                        document_objects = Document(
                            page_content=table_content,
                            metadata={
                                "source": title,
                                "document_id": document.get("document_id", "default_id"),
                                "chunk_number": "table"
                            }
                        )
                        combined_document_objects.append(document_objects)
        
        # Chunk the rest of the text
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=model_run_state.chunk_size, chunk_overlap=model_run_state.chunk_overlap)
        chunks_document = text_splitter.split_text(text)

        for chunk_number, chunk in enumerate(chunks_document, start=1):
            document_objects = Document(
                page_content=chunk,
                metadata={
                    "source": title,
                    "document_id": document.get("document_id", "default_id"),
                    "chunk_number": chunk_number
                }
            )
            combined_document_objects.append(document_objects)

    return combined_document_objects


def chunk_with_section_based(app_state, model_run_state):
    combined_document_objects = []
    dprint(app_state, "Using Section-Based Chunking for documents.")

    for document in app_state.documents:
        text = document["single_text_document"]
        title = document["title"]

        # Split the text by headings
        sections = re.split(r"\n[A-Z].+?\n", text)

        # Chunk each section
        text_splitter = RecursiveCharacterTextSplitter(chunk_size=model_run_state.chunk_size, chunk_overlap=model_run_state.chunk_overlap)
        for section_number, section in enumerate(sections, start=1):
            chunks_document = text_splitter.split_text(section)
            for chunk_number, chunk in enumerate(chunks_document, start=1):
                document_objects = Document(
                    page_content=chunk,
                    metadata={
                        "source": title,
                        "document_id": document.get("document_id", "default_id"),
                        "section_number": section_number,
                        "chunk_number": chunk_number
                    }
                )
                combined_document_objects.append(document_objects)

    return combined_document_objects


def chunk_with_semantic_splitter(app_state, model_run_state):
    # Load pre-trained model for embeddings
    model = SentenceTransformer('all-MiniLM-L6-v2')

    combined_document_objects = []
    dprint(app_state, "Using Semantic-Based Chunking for documents.")

    for document in app_state.documents:
        text = document["single_text_document"]
        title = document["title"]

        # Split text into sentences or paragraphs
        sentences = text.split(". ")  # Simple split by sentence (you can refine this)
        sentence_embeddings = model.encode(sentences)

        # Group sentences into chunks based on semantic similarity
        chunks = []
        current_chunk = []
        for i in range(len(sentences) - 1):
            current_chunk.append(sentences[i])
            # Calculate similarity between consecutive sentences
            sim = cosine_similarity([sentence_embeddings[i]], [sentence_embeddings[i + 1]])[0][0]
            if sim < 0.7 or len(current_chunk) >= model_run_state.chunk_size:
                # If similarity is below threshold or chunk size is reached, start a new chunk
                chunks.append(" ".join(current_chunk))
                current_chunk = []

        # Add the final chunk
        if current_chunk:
            chunks.append(" ".join(current_chunk))

        # Create document objects for the chunks
        for chunk_number, chunk in enumerate(chunks, start=1):
            document_objects = Document(
                page_content=chunk,
                metadata={
                    "source": title,
                    "document_id": document.get("document_id", "default_id"),
                    "chunk_number": chunk_number
                }
            )
            combined_document_objects.append(document_objects)

    return combined_document_objects