fintorch 0.6.0

AI4FinTech project repository

AI4FinTech project repository

• Free software: MIT license

• Documentation: https://fintorch.readthedocs.io.

FinTorch - Machine Learning for FinTech

The integration of AI in the financial sector demands specialized tools that can handle the unique challenges of this field, especially in regulatory compliance and risk management. Building on the familiarity and robustness of PyTorch, FinTorch aims to bridge the gap between AI technology and the financial industry needs.

Goal

Develop FinTorch, an open-source machine learning library as an extension of PyTorch, specifically tailored for the FinTech industry’s compliance and risk management requirements.

Key Objectives

1. Specialized Financial AI Models Implement state-of-the-art machine learning models for financial data analysis, fraud detection, risk assessment, and regulatory compliance, seamlessly integrating with PyTorch’s existing framework.

2. Regulatory Compliance Toolkit Provide tools specifically designed for monitoring and ensuring adherence to financial regulations using AI.

3. User-Friendly API Maintain a tensor-centric API, consistent with PyTorch, ensuring ease of use for those familiar with PyTorch. Aim for simplicity, where basic models can be implemented in as few as 10-20 lines of code.

4. Extensibility for Research Offer a flexible platform for academic and industry researchers to develop and test new AI models for FinTech, with support for custom architectures and novel strategies.

5. Scalability and Real-World Application Focus on scalability to handle large-scale financial data and real-world scenarios.

6. Ethical and Responsible AI Practices Embed principles of sustainable and responsible AI, ensuring that models adhere to ethical standards and contribute positively to the FinTech ecosystem.

7. Educational Resources and Community Support Provide comprehensive documentation, tutorials, and masterclasses to facilitate learning and collaboration within the AI4FinTech community.

Impact

FinTorch will not only streamline the process of regulatory compliance for FinTech companies but also foster innovation and research in AI-driven financial technologies. It will serve as a crucial tool for industry professionals, researchers, and government institutions, aligning with the AI4FinTech community’s objectives of knowledge dissemination and development of responsible, cutting-edge financial solutions.

Getting started

Please install the package as follows:

pip install fintorch

Required Dependencies

• The following instalation commands for the required dependencies use custom index URLs provided by PyTorch Geometric (PyG).

To determine TORCH version and DEVICE type run:

python -c "import torch; print('TORCH:',torch.__version__,'\nDEVICE:', 'cu'+torch.version.cuda.replace('.','')) if torch.cuda.is_available() else print('TORCH:',str(torch.__version__).split('+')[0],'\nDEVICE: cpu')"

For Linux:

Set TORCH and DEVICE:

export TORCH={your_pytorch_version}
export DEVICE={your_device}

Or, in the following urls, replace ${TORCH} and ${DEVICE} with the appropriate version numbers for your environment (e.g., with CUDA: “1.12.0” and “cu113” or with CPU: “1.12.0” and “cpu”).

Download the required dependencies using the following PyG installation commands:

pip install pyg-lib -f https://data.pyg.org/whl/torch-${TORCH}+${DEVICE}.html
pip install torch-scatter torch-sparse -f https://data.pyg.org/whl/torch-${TORCH}+${DEVICE}.html

For Windows:

Set TORCH and DEVICE:

set "TORCH=your_pytorch_version"
set "DEVICE=your_device"

Or, in the following urls, replace %TORCH% and %DEVICE% with the appropriate version numbers for your environment (e.g., with CUDA: “1.12.0” and “cu113” or with CPU: “1.12.0” and “cpu”).

Download the required dependencies using the following PyG installation commands:

pip install pyg-lib -f https://data.pyg.org/whl/torch-%TORCH%+%DEVICE%.html
pip install torch-scatter torch-sparse -f https://data.pyg.org/whl/torch-%TORCH%+%DEVICE%.html

Description of the Structure

• fintorch Directory: Contains the core library modules.

  • models: Core models for compliance monitoring, fraud detection, risk assessment, and sustainable finance.

  • datasets: Financial datasets and data processing utilities.

  • utils: Helper tools and functions for compliance and other financial applications.

  • training: Training and evaluation scripts for the models.

• examples Directory: Example scripts demonstrating the use of FinTorch in different scenarios.

• tests Directory: Unit and integration tests for the library.

• benchmarks Directory: Benchmark scripts and resources for testing the performance of the library.

• docs Directory: Documentation files, including build scripts and source files.

• docker Directory: Dockerfile and related resources for containerizing the FinTorch library.

• conda Directory: Scripts and files needed for building a Conda package of the library.

• tutorials Directory: Jupyter notebooks that provide tutorials on how to use the library for various FinTech applications.