https://github.com/zhampel/clustergan

Pytorch Implementation of ClusterGAN (arXiv:1809.03627)
https://github.com/zhampel/clustergan

Last synced: about 4 hours ago
JSON representation

Pytorch Implementation of ClusterGAN (arXiv:1809.03627)

• Host: GitHub
• URL: https://github.com/zhampel/clustergan
• Owner: zhampel
• License: mit
• Created: 2019-04-16T00:18:28.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2023-07-06T21:26:28.000Z (about 1 year ago)
• Last Synced: 2024-04-24T12:13:47.999Z (5 months ago)
• Language: Python
• Size: 1.25 MB
• Stars: 88
• Watchers: 6
• Forks: 22
• Open Issues: 7
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# ClusterGAN: A PyTorch Implementation

This is a PyTorch implementation of [ClusterGAN](https://arxiv.org/abs/1809.03627),

an approach to unsupervised clustering using generative adversarial networks.

## Requirements

The package as well as the necessary requirements can be installed by running `make` or via

```

virtualenv -p /usr/local/bin/python3 venv

source venv/bin/activate

python setup.py install

```

## Run ClusterGAN on MNIST

To run ClusterGAN on the MNIST dataset, ensure the package is setup and then run

```

python train.py -r test_run -s mnist -b 64 -n 300

```

where a directory `runs/mnist/test_run` will be made and contain the generated output

(models, example generated instances, training figures) from the training run.

The `-r` option denotes the run name, `-s` the dataset (currently MNIST and Fashion-MNIST),

`-b` the batch size, and `-n` the number of training epochs.

Below is an example set of training curves for 200 epochs, batch size of 64 on the MNIST dataset.



  

  



## Generated Examples

To generate examples from randomly sampled latent space variables, 

```

python gen-examples -r test_run -s mnist -b 100

```

Here are some example generated images by specified class (each row) of the learned labels in latent space.



  



## TSNE Figure

To produce a TSNE figure depicting the clustering of the latent space encoding of real images,

```

python tsne-cluster.py -r test_run -s mnist

```

Below is the tSNE clustering figure of the latent space vectors of true MNIST images fed into the encoder.



  



## License

[MIT License](LICENSE)

Copyright (c) 2018 Zigfried Hampel-Arias