pytorch-tabular/pytorch_tabular
A unified framework for Deep Learning Models on tabular data
What it solves
PyTorch Tabular provides a unified, high-level API to apply deep learning architectures to tabular data, making these complex models more accessible for both real-world applications and academic research. It removes the friction of manually implementing various tabular-specific neural networks.
How it works
Built on PyTorch and PyTorch Lightning, the library provides a standardized interface for training, evaluating, and predicting. Users define a DataConfig for their columns, a ModelConfig for the specific architecture, and a TrainerConfig for the learning process. This allows for easy switching between different deep learning models for tabular data without rewriting the rest of the pipeline.
Who it’s for
Data scientists and ML researchers who want to use deep learning for tabular datasets instead of traditional gradient boosting machines, or those who looking to quickly prototype and benchmark different neural network architectures for structured data.
Highlights
- Wide Model Selection: Includes a variety of architectures such as TabNet, TabTransformer, FT Transformer, NODE, and GANDALF.
- Scalability: Leverages PyTorch Lightning to scale training across CPUs and GPUs with automatic logging.
- Customization: Supports the implementation of custom models through a dedicated tutorial and framework.
- Semi-Supervised Learning: Includes a Denoising AutoEncoder for learning robust feature representations from noisy data.