https://github.com/paganpasta/onmixup

NeurIPS Reproducbility Challenge 2019
https://github.com/paganpasta/onmixup

Last synced: 3 months ago
JSON representation

NeurIPS Reproducbility Challenge 2019

• Host: GitHub
• URL: https://github.com/paganpasta/onmixup
• Owner: paganpasta
• Created: 2019-12-27T22:03:42.000Z (almost 5 years ago)
• Default Branch: master
• Last Pushed: 2020-02-25T16:19:04.000Z (over 4 years ago)
• Last Synced: 2024-05-12T22:53:07.537Z (6 months ago)
• Language: Jupyter Notebook
• Size: 3.68 MB
• Stars: 9
• Watchers: 2
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-Mixup - [Code

README

        
``` NeurIPS 2019 Reproducibility Project ```

Paper reproduced: [pdf](https://arxiv.org/pdf/1905.11001.pdf)

#### Getting Started

#### Pre-requisite

Install [NVIDIA Docker](https://github.com/NVIDIA/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.