Welcome to the PyTorch Lightning implementation of VIME: Extending the Success of Self- and Semi-supervised Learning to Tabular Domain (ICML'2020). This project is an unofficial implementation and was developed to mirror the methods described in the VIME paper as closely as possible.
pip install -r requirements.txt
See the example_*.ipynb
Difference between the official tensorflow implementation
The official implementation generates static mask vectors, feature vectors, and corrupted samples during initialization of the dataset.
However, we generate them dynamically whenever call '__getitem__' of the dataset to avoid bias.