ContrastiveMixup Pytorch

This repository contains the code for (arxiv link)

@misc{darabi2021contrastive,
      title={Contrastive Mixup: Self- and Semi-Supervised learning for Tabular Domain}, 
      author={Sajad Darabi and Shayan Fazeli and Ali Pazoki and Sriram Sankararaman and Majid Sarrafzadeh},
      year={2021},
      eprint={2108.12296},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Dependencies

The project was run on a conda virtual environment on Ubuntu 18.04.5 LTS.

Checkout the requirements.txt file, if you have conda pre-installed cd into the directory where you have downloaded the source code and run the following

conda create -n contrastivemixup python==3.7
conda activate contrastivemixup

pip install -r requirements.txt

Running Experiments

To run experiments they are launched from the train.py file. For example, to run ContrastiveMixup on MNIST use the following command

python train.py -c ./configs/mnist/contrastivemixup.json --pretrain

The trainer, data loader, model, optimizer, settings are all specified in the ./configs/mnist/contrastivemixup.json file. The --pretrain options specifies whether to run the pretraining phase (i.e. training the encoder).

TODO

• [ ] Add instructions to run code
• [ ] File structure
• [ ] Config file instructions
• [ ] Comment/clean code
• [ ] Jupyter lab demo