A PyTorch extension for building and training deep learning models on tabular data with mixed column types, including text and images.
PyTorch Frame is a deep learning extension for PyTorch aimed at heterogeneous tabular data, where columns can be numerical, categorical, multi-categorical, timestamp, text, image, or embedding values. It provides a modular pipeline that turns a pandas DataFrame into a tensor frame, encodes each column type, models interactions between columns, and decodes predictions per row.
The library includes implementations of several deep tabular architectures, such as FTTransformer, ResNet, TabNet, TabTransformer, Trompt and ExcelFormer, along with XGBoost, CatBoost and LightGBM examples that use Optuna for hyperparameter tuning. It also ships with readily usable benchmark datasets and supports custom data. Text columns can be encoded with embeddings from OpenAI, Cohere, Voyage AI or Hugging Face models.
Because it builds directly on PyTorch and works with PyTorch Geometric, it can be combined with other PyTorch models, for example in relational database learning.
Summary generated by AI from the project's README. GitHub stats refresh daily.
A Python library for self-supervised learning on images, offering PyTorch building blocks for training vision models without labels.
AutoML library that trains and ensembles ML models for tabular, time series, and multimodal data with a few lines of Python.
TensorBoard is a web-based visualization toolkit for inspecting TensorFlow training runs, metrics, graphs and embeddings.
A Python NLP framework built on PyTorch for named entity recognition, sentiment analysis, PoS tagging, and text embeddings.
Model-definition framework for state-of-the-art text, vision, audio and multimodal machine learning models, for inference and training.
Keras 3 is a deep learning framework that runs on JAX, TensorFlow, PyTorch, or OpenVINO for inference.