vae_oversampler

vae_oversampler provides an API similar to imblearn to oversample a minority class of a dataset. Under the hood it uses keras to build a variational autoencoder that learns the underlying data probability distribution and then samples from that distribution to generate synthetic minority examples.

Tech

vae_oversampler uses a number of open source projects to work properly:

• [keras] - for deep learning- to build the variational autoencoder
• [sklearn] - primarily to standard scale your data (optional)
• [numpy] - numerical methods

And of course vae_oversampler itself is open source with a public repository on GitHub.

Installation

vae_oversampler requires keras to run.

Install the dependencies and install using pip

$ pip install vae_oversampler

Todos

• Write Tests
• Comply with PEP8
• Better error handling
• Add more options for how many samples to generate
• get travis working properly (with TensorFlow)

License

MIT