tungtumkit / onmixup

NeurIPS Reproducbility Challenge 2019

NeurIPS 2019 Reproducibility Project

Paper reproduced: pdf

Getting Started

Pre-requisite

Install NVIDIA Docker

Building & Running the docker image

1. nvidia-docker build --rm --tag on_mixup:latest .
2. nvidia-docker run -p 8000:8000 -v $PATH/TO/PROJECT_DIR:/home/on_mixup --rm --name on_mixup -it on_mixup:latest

Generating The Density Plots

Firstly, start the jupyter notebook

jupyter notebook --ip=0.0.0.0 --port=8000 --allow-root --NotebookApp.token='' --NotebookApp.password=''

There are two notebooks corrsponding to scenarios of Mixup and no Mixup. Details on how to run them are provided in the notebooks itself.

Training

There are files named:

1. cifar.py
2. fmnist.py
3. stl_10.py

To the train the neural network on the particular dataset execute the corresponding script, for example, python cifar.py

Evaluation

evaluation.py takes as an argument a config.yaml file. Sample configs are in the eval_config folder. To run an evaluation on the dataset, execute the following command python evaluation.py --config eval_config/cifar.yaml The fields of the config are self explanatory.