A Variational Auto-Encoder library for PyTorch with torchbearer
Torchbearer.variational is a companion package to torchbearer which is intended to re-implement state of the art models and practices relating to the world of Variational Auto-Encoders (VAEs). The goal is to provide everything from useful abstractions to complete re-implementations of papers. This is in order to support both research and teaching / learning regarding VAEs.
The easiest way to install visual is with pip:
pip install torchbearer-variational
Alternatively, build from source with:
pip install git+https://github.com/pytorchbearer/variational
Currently, variational only includes abstractions for simple VAEs and some accompaniments, the next steps are as follows:
- Construct some separate part of the docs for the variational content
- Implement a series of standard models with associated notes pages and example usages
- Implement other divergences not in PyTorch such as MMD, Jensen-Shannon, etc.
- Implement and document tools for sampling the latent spaces of models and producing figures
- Implement other dataloaders not in torchvision and add associated docs