VAEGAN

Module for a GAN sampling from a Variational Autoencoder (VAE).

The VAE is inspired by the implementation of "Survival-oriented embeddings with application to CT scans of colorectal carcinoma patients with liver metastases" by Tobias Weber, 2021. The basic structure of the underlying AE-GAN architecture is inspired by ecgan.modules.generative.aegan.AEGAN.

class ecgan.modules.generative.vaegan.VAEGAN(cfg, seq_len, num_channels)[source]

Bases: ecgan.modules.generative.aegan.AEGAN

Variational Autoencoder for encoding data to latent space, reconstructing it and judging the quality.

property generator_sampler: ecgan.utils.sampler.VAEGANGeneratorSampler

Return the sampler used to sample from the generator.

Return type: VAEGANGeneratorSampler

property criterion_gen: ecgan.utils.losses.VAEGANGeneratorLoss

Return the criterion for the generator.

Return type: VAEGANGeneratorLoss

get_kl_weight()[source]

Get annealed weight for Kullback-Leibler Divergence.

Return type: float

static configure()[source]

Return the default configuration for the VAEGAN model.

Return type: Dict