oceansweep's picture
?
ed28876
raw
history blame
19.1 kB
# Chunk_Lib.py
#########################################
# Chunking Library
# This library is used to perform chunking of input files.
# Currently, uses naive approaches. Nothing fancy.
#
####
# Import necessary libraries
import logging
import re
from typing import List, Optional, Tuple, Dict, Any
from openai import OpenAI
from tqdm import tqdm
#
# Import 3rd party
from transformers import GPT2Tokenizer
import nltk
from nltk.tokenize import sent_tokenize, word_tokenize
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.metrics.pairwise import cosine_similarity
#
# Import Local
from App_Function_Libraries.Tokenization_Methods_Lib import openai_tokenize
from App_Function_Libraries.Utils import load_comprehensive_config
#
#######################################################################################################################
# Function Definitions
#
# FIXME - Make sure it only downloads if it already exists, and does a check first.
# Ensure NLTK data is downloaded
def ntlk_prep():
nltk.download('punkt')
# Load GPT2 tokenizer
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
# Load Config file for API keys
config = load_comprehensive_config()
openai_api_key = config.get('API', 'openai_api_key', fallback=None)
def load_document(file_path):
with open(file_path, 'r') as file:
text = file.read()
return re.sub('\\s+', ' ', text).strip()
def improved_chunking_process(text: str, chunk_options: Dict[str, Any]) -> List[Dict[str, Any]]:
chunk_method = chunk_options.get('method', 'words')
max_chunk_size = chunk_options.get('max_size', 300)
overlap = chunk_options.get('overlap', 0)
language = chunk_options.get('language', 'english')
adaptive = chunk_options.get('adaptive', False)
multi_level = chunk_options.get('multi_level', False)
if adaptive:
max_chunk_size = adaptive_chunk_size(text, max_chunk_size)
if multi_level:
chunks = multi_level_chunking(text, chunk_method, max_chunk_size, overlap, language)
else:
if chunk_method == 'words':
chunks = chunk_text_by_words(text, max_chunk_size, overlap)
elif chunk_method == 'sentences':
chunks = chunk_text_by_sentences(text, max_chunk_size, overlap, language)
elif chunk_method == 'paragraphs':
chunks = chunk_text_by_paragraphs(text, max_chunk_size, overlap)
elif chunk_method == 'tokens':
chunks = chunk_text_by_tokens(text, max_chunk_size, overlap)
else:
chunks = [text] # No chunking applied
return [{'text': chunk, 'metadata': get_chunk_metadata(chunk, text)} for chunk in chunks]
def adaptive_chunk_size(text: str, base_size: int) -> int:
# Simple adaptive logic: adjust chunk size based on text complexity
avg_word_length = sum(len(word) for word in text.split()) / len(text.split())
if avg_word_length > 6: # Arbitrary threshold for "complex" text
return int(base_size * 0.8) # Reduce chunk size for complex text
return base_size
def multi_level_chunking(text: str, method: str, max_size: int, overlap: int, language: str) -> List[str]:
# First level: chunk by paragraphs
paragraphs = chunk_text_by_paragraphs(text, max_size * 2, overlap)
# Second level: chunk each paragraph further
chunks = []
for para in paragraphs:
if method == 'words':
chunks.extend(chunk_text_by_words(para, max_size, overlap))
elif method == 'sentences':
chunks.extend(chunk_text_by_sentences(para, max_size, overlap, language))
else:
chunks.append(para)
return chunks
def chunk_text_by_words(text: str, max_words: int = 300, overlap: int = 0) -> List[str]:
words = text.split()
chunks = []
for i in range(0, len(words), max_words - overlap):
chunk = ' '.join(words[i:i + max_words])
chunks.append(chunk)
return post_process_chunks(chunks)
def chunk_text_by_sentences(text: str, max_sentences: int = 10, overlap: int = 0, language: str = 'english') -> List[
str]:
nltk.download('punkt', quiet=True)
sentences = nltk.sent_tokenize(text, language=language)
chunks = []
for i in range(0, len(sentences), max_sentences - overlap):
chunk = ' '.join(sentences[i:i + max_sentences])
chunks.append(chunk)
return post_process_chunks(chunks)
def chunk_text_by_paragraphs(text: str, max_paragraphs: int = 5, overlap: int = 0) -> List[str]:
paragraphs = re.split(r'\n\s*\n', text)
chunks = []
for i in range(0, len(paragraphs), max_paragraphs - overlap):
chunk = '\n\n'.join(paragraphs[i:i + max_paragraphs])
chunks.append(chunk)
return post_process_chunks(chunks)
def chunk_text_by_tokens(text: str, max_tokens: int = 1000, overlap: int = 0) -> List[str]:
# This is a simplified token-based chunking. For more accurate tokenization,
# consider using a proper tokenizer like GPT-2 TokenizerFast
words = text.split()
chunks = []
current_chunk = []
current_token_count = 0
for word in words:
word_token_count = len(word) // 4 + 1 # Rough estimate of token count
if current_token_count + word_token_count > max_tokens and current_chunk:
chunks.append(' '.join(current_chunk))
current_chunk = current_chunk[-overlap:] if overlap > 0 else []
current_token_count = sum(len(w) // 4 + 1 for w in current_chunk)
current_chunk.append(word)
current_token_count += word_token_count
if current_chunk:
chunks.append(' '.join(current_chunk))
return post_process_chunks(chunks)
def post_process_chunks(chunks: List[str]) -> List[str]:
return [chunk.strip() for chunk in chunks if chunk.strip()]
def get_chunk_metadata(chunk: str, full_text: str) -> Dict[str, Any]:
start_index = full_text.index(chunk)
return {
'start_index': start_index,
'end_index': start_index + len(chunk),
'word_count': len(chunk.split()),
'char_count': len(chunk)
}
# Hybrid approach, chunk each sentence while ensuring total token size does not exceed a maximum number
def chunk_text_hybrid(text, max_tokens=1000):
sentences = nltk.tokenize.sent_tokenize(text)
chunks = []
current_chunk = []
current_length = 0
for sentence in sentences:
tokens = tokenizer.encode(sentence)
if current_length + len(tokens) <= max_tokens:
current_chunk.append(sentence)
current_length += len(tokens)
else:
chunks.append(' '.join(current_chunk))
current_chunk = [sentence]
current_length = len(tokens)
if current_chunk:
chunks.append(' '.join(current_chunk))
return chunks
# Thanks openai
def chunk_on_delimiter(input_string: str,
max_tokens: int,
delimiter: str) -> List[str]:
chunks = input_string.split(delimiter)
combined_chunks, _, dropped_chunk_count = combine_chunks_with_no_minimum(
chunks, max_tokens, chunk_delimiter=delimiter, add_ellipsis_for_overflow=True)
if dropped_chunk_count > 0:
print(f"Warning: {dropped_chunk_count} chunks were dropped due to exceeding the token limit.")
combined_chunks = [f"{chunk}{delimiter}" for chunk in combined_chunks]
return combined_chunks
def recursive_summarize_chunks(chunks, summarize_func, custom_prompt):
summarized_chunks = []
current_summary = ""
for i, chunk in enumerate(chunks):
if i == 0:
current_summary = summarize_func(chunk, custom_prompt)
else:
combined_text = current_summary + "\n\n" + chunk
current_summary = summarize_func(combined_text, custom_prompt)
summarized_chunks.append(current_summary)
return summarized_chunks
# Sample text for testing
sample_text = """
Natural language processing (NLP) is a subfield of linguistics, computer science, and artificial intelligence
concerned with the interactions between computers and human language, in particular how to program computers
to process and analyze large amounts of natural language data. The result is a computer capable of "understanding"
the contents of documents, including the contextual nuances of the language within them. The technology can then
accurately extract information and insights contained in the documents as well as categorize and organize the documents themselves.
Challenges in natural language processing frequently involve speech recognition, natural language understanding,
and natural language generation.
Natural language processing has its roots in the 1950s. Already in 1950, Alan Turing published an article titled
"Computing Machinery and Intelligence" which proposed what is now called the Turing test as a criterion of intelligence.
"""
# Example usage of different chunking methods
# print("Chunking by words:")
# print(chunk_text_by_words(sample_text, max_words=50))
#
# print("\nChunking by sentences:")
# print(chunk_text_by_sentences(sample_text, max_sentences=2))
#
# print("\nChunking by paragraphs:")
# print(chunk_text_by_paragraphs(sample_text, max_paragraphs=1))
#
# print("\nChunking by tokens:")
# print(chunk_text_by_tokens(sample_text, max_tokens=50))
#
# print("\nHybrid chunking:")
# print(chunk_text_hybrid(sample_text, max_tokens=50))
#######################################################################################################################
#
# Experimental Semantic Chunking
#
# Chunk text into segments based on semantic similarity
def count_units(text, unit='tokens'):
if unit == 'words':
return len(text.split())
elif unit == 'tokens':
return len(word_tokenize(text))
elif unit == 'characters':
return len(text)
else:
raise ValueError("Invalid unit. Choose 'words', 'tokens', or 'characters'.")
def semantic_chunking(text, max_chunk_size=2000, unit='words'):
nltk.download('punkt', quiet=True)
sentences = sent_tokenize(text)
vectorizer = TfidfVectorizer()
sentence_vectors = vectorizer.fit_transform(sentences)
chunks = []
current_chunk = []
current_size = 0
for i, sentence in enumerate(sentences):
sentence_size = count_units(sentence, unit)
if current_size + sentence_size > max_chunk_size and current_chunk:
chunks.append(' '.join(current_chunk))
overlap_size = count_units(' '.join(current_chunk[-3:]), unit) # Use last 3 sentences for overlap
current_chunk = current_chunk[-3:] # Keep last 3 sentences for overlap
current_size = overlap_size
current_chunk.append(sentence)
current_size += sentence_size
if i + 1 < len(sentences):
current_vector = sentence_vectors[i]
next_vector = sentence_vectors[i + 1]
similarity = cosine_similarity(current_vector, next_vector)[0][0]
if similarity < 0.5 and current_size >= max_chunk_size // 2:
chunks.append(' '.join(current_chunk))
overlap_size = count_units(' '.join(current_chunk[-3:]), unit)
current_chunk = current_chunk[-3:]
current_size = overlap_size
if current_chunk:
chunks.append(' '.join(current_chunk))
return chunks
def semantic_chunk_long_file(file_path, max_chunk_size=1000, overlap=100):
try:
with open(file_path, 'r', encoding='utf-8') as file:
content = file.read()
chunks = semantic_chunking(content, max_chunk_size, overlap)
return chunks
except Exception as e:
logging.error(f"Error chunking text file: {str(e)}")
return None
#######################################################################################################################
#######################################################################################################################
#
# OpenAI Rolling Summarization
#
client = OpenAI(api_key=openai_api_key)
def get_chat_completion(messages, model='gpt-4-turbo'):
response = client.chat.completions.create(
model=model,
messages=messages,
temperature=0,
)
return response.choices[0].message.content
# This function combines text chunks into larger blocks without exceeding a specified token count.
# It returns the combined chunks, their original indices, and the number of dropped chunks due to overflow.
def combine_chunks_with_no_minimum(
chunks: List[str],
max_tokens: int,
chunk_delimiter="\n\n",
header: Optional[str] = None,
add_ellipsis_for_overflow=False,
) -> Tuple[List[str], List[int]]:
dropped_chunk_count = 0
output = [] # list to hold the final combined chunks
output_indices = [] # list to hold the indices of the final combined chunks
candidate = (
[] if header is None else [header]
) # list to hold the current combined chunk candidate
candidate_indices = []
for chunk_i, chunk in enumerate(chunks):
chunk_with_header = [chunk] if header is None else [header, chunk]
# FIXME MAKE NOT OPENAI SPECIFIC
if len(openai_tokenize(chunk_delimiter.join(chunk_with_header))) > max_tokens:
print(f"warning: chunk overflow")
if (
add_ellipsis_for_overflow
# FIXME MAKE NOT OPENAI SPECIFIC
and len(openai_tokenize(chunk_delimiter.join(candidate + ["..."]))) <= max_tokens
):
candidate.append("...")
dropped_chunk_count += 1
continue # this case would break downstream assumptions
# estimate token count with the current chunk added
# FIXME MAKE NOT OPENAI SPECIFIC
extended_candidate_token_count = len(openai_tokenize(chunk_delimiter.join(candidate + [chunk])))
# If the token count exceeds max_tokens, add the current candidate to output and start a new candidate
if extended_candidate_token_count > max_tokens:
output.append(chunk_delimiter.join(candidate))
output_indices.append(candidate_indices)
candidate = chunk_with_header # re-initialize candidate
candidate_indices = [chunk_i]
# otherwise keep extending the candidate
else:
candidate.append(chunk)
candidate_indices.append(chunk_i)
# add the remaining candidate to output if it's not empty
if (header is not None and len(candidate) > 1) or (header is None and len(candidate) > 0):
output.append(chunk_delimiter.join(candidate))
output_indices.append(candidate_indices)
return output, output_indices, dropped_chunk_count
def rolling_summarize(text: str,
detail: float = 0,
model: str = 'gpt-4-turbo',
additional_instructions: Optional[str] = None,
minimum_chunk_size: Optional[int] = 500,
chunk_delimiter: str = ".",
summarize_recursively=False,
verbose=False):
"""
Summarizes a given text by splitting it into chunks, each of which is summarized individually.
The level of detail in the summary can be adjusted, and the process can optionally be made recursive.
Parameters:
- text (str): The text to be summarized.
- detail (float, optional): A value between 0 and 1
indicating the desired level of detail in the summary. 0 leads to a higher level summary, and 1 results in a more
detailed summary. Defaults to 0.
- additional_instructions (Optional[str], optional): Additional instructions to provide to the
model for customizing summaries. - minimum_chunk_size (Optional[int], optional): The minimum size for text
chunks. Defaults to 500.
- chunk_delimiter (str, optional): The delimiter used to split the text into chunks. Defaults to ".".
- summarize_recursively (bool, optional): If True, summaries are generated recursively, using previous summaries for context.
- verbose (bool, optional): If True, prints detailed information about the chunking process.
Returns:
- str: The final compiled summary of the text.
The function first determines the number of chunks by interpolating between a minimum and a maximum chunk count
based on the `detail` parameter. It then splits the text into chunks and summarizes each chunk. If
`summarize_recursively` is True, each summary is based on the previous summaries, adding more context to the
summarization process. The function returns a compiled summary of all chunks.
"""
# check detail is set correctly
assert 0 <= detail <= 1
# interpolate the number of chunks based to get specified level of detail
max_chunks = len(chunk_on_delimiter(text, minimum_chunk_size, chunk_delimiter))
min_chunks = 1
num_chunks = int(min_chunks + detail * (max_chunks - min_chunks))
# adjust chunk_size based on interpolated number of chunks
# FIXME MAKE NOT OPENAI SPECIFIC
document_length = len(openai_tokenize(text))
chunk_size = max(minimum_chunk_size, document_length // num_chunks)
text_chunks = chunk_on_delimiter(text, chunk_size, chunk_delimiter)
if verbose:
print(f"Splitting the text into {len(text_chunks)} chunks to be summarized.")
# FIXME MAKE NOT OPENAI SPECIFIC
print(f"Chunk lengths are {[len(openai_tokenize(x)) for x in text_chunks]}")
# set system message - FIXME
system_message_content = "Rewrite this text in summarized form."
if additional_instructions is not None:
system_message_content += f"\n\n{additional_instructions}"
accumulated_summaries = []
for i, chunk in enumerate(tqdm(text_chunks)):
if summarize_recursively and accumulated_summaries:
# Combine previous summary with current chunk for recursive summarization
combined_text = accumulated_summaries[-1] + "\n\n" + chunk
user_message_content = f"Previous summary and new content to summarize:\n\n{combined_text}"
else:
user_message_content = chunk
messages = [
{"role": "system", "content": system_message_content},
{"role": "user", "content": user_message_content}
]
response = get_chat_completion(messages, model=model)
accumulated_summaries.append(response)
final_summary = '\n\n'.join(accumulated_summaries)
return final_summary