tabulens 0.1.0

LLM-powered PDF table extractor

Tabulens

Tabulens is a Python package that intelligently extracts and restructures tables from PDF files using advanced computer vision and Large Language Models (LLMs). It automatically detects table structures, manages complex hierarchical tables, and exports data into structured formats like CSV and pandas DataFrames.

Features

• 🔍 Automatic Table Detection: Uses computer vision to identify table regions.
• 🧠 Intelligent Restructuring: Leverages LLMs to understand and restructure hierarchical tables.
• 📊 Multiple Output Formats: Supports CSV and pandas DataFrame outputs.
• 🎯 High Accuracy: Combines computer vision preprocessing with LLM analysis for robust extraction.
• 🔧 Flexible Models: Supports both OpenAI GPT and Google Gemini models.
• 📝 Hierarchy Preservation: Flattens nested tables while maintaining parent-child relationships.
• 🚀 Easy to Use: Simple API and command-line interface.

Installation

From PyPI:

pip install tabulens

Or directly from GitHub:

pip install git+https://github.com/astonishedrobo/tabulens.git

Quick Start

Python API

from tabulens import TableExtractor

extractor = TableExtractor(
    model_name='gpt:gpt-4o-mini', # gemini:gemini-2.0-flash
    temperature=0.7
)

dfs = extractor.extract_tables(
    file_path='path/to/document.pdf',
    save=True,
    max_tries=3,
    print_logs=True
)

for i, df in enumerate(dataframes):
    if df is not None:
        print(f"Table {i+1}")
        print(df.head())

Command Line Interface

# OpenAI 
tabulens --pdf path/to/document.pdf --model gpt:gpt-4o-mini --temperature 0.7 --max_tries 3 --log

# Gemini
tabulens --pdf path/to/document.pdf --model gemini:gemini-2.0-flash --temperature 0.7 --max_tries 3 --log

CLI Options

• --pdf: Path to the PDF file (required)
• --model: Model name (gpt:gpt-4o-mini, gemini:gemini-2.0-flash, gpt:gpt-4o, gemini:gemini-2.5-flash-preview-05-20, etc.) [default: gpt:gpt-4o-mini]. For OpenAI models, use the prefix gpt:, and for Gemini models, use the prefix gemini:. ⚠️ Make sure to select models that support image inputs. You can use any of the mentioned examples for convenience.
• --temperature: Generation temperature (0.0-1.0) [default: 0.7]
• --max_tries: Maximum retries per table extraction [default: 3] [Increase this value to enhance accuracy, as more attempts allow the system additional opportunities to correctly extract tables.]
• --log: Print detailed logs

Environment Variable Setup

Before running the program, set the required API environment variables.

For CLI usage:

export OPENAI_API_KEY=<your_openai_api_key>
export GOOGLE_API_KEY=<your_google_api_key>

For Python API usage, load environment variables using python-dotenv:

from dotenv import load_dotenv
load_dotenv("path/to/.env")

Credits

Tabulens depends on these excellent open-source projects:

• LangChain
• OpenCV
• NumPy
• Pandas
• pdf2image
• tqdm

License

This project is licensed under the MIT License.