Loss functions

Implementation of various loss functions for PyTorch models.

class ecgan.utils.losses.SupervisedLoss[source]

Bases: ecgan.utils.configurable.Configurable

Base class for supervised loss functions.

forward(input_tensor, target_tensor)[source]

Return the loss of a given component.

Return type: Tensor

static configure()[source]

Return the default configuration for a general loss function.

Return type: Dict

class ecgan.utils.losses.L2Loss(reduction='mean')[source]

Bases: ecgan.utils.losses.SupervisedLoss

Wrapper over the mean squared error loss of the torch module.

static configure()[source]

Return the default configuration for a general loss function.

Return type: Dict

class ecgan.utils.losses.BCELoss(reduction='mean')[source]

Bases: ecgan.utils.losses.SupervisedLoss

Wrapper over the binary cross entropy loss of the torch module.

static configure()[source]

Return the default configuration for a general loss function.

Return type: Dict

class ecgan.utils.losses.CrossEntropyLoss(reduction='mean')[source]

Bases: ecgan.utils.losses.SupervisedLoss

Wrapper over the cross entropy loss of the torch module.

static configure()[source]

Return the default configuration for a general loss function.

Return type: Dict

class ecgan.utils.losses.KLLoss[source]

Bases: object

Kullback-Divergence loss for the usage in variational auto-encoders.

static forward(mean_value, log_var)[source]

Calculate Kullback-Leibler divergence for standard Gaussian distribution.

Calculate KL divergence for a given expected value and log variance. The input tensors are expected to be in shape (N x DIM) where N is the number of samples and DIM is the dimension of the multivariate Gaussian. The result will be the average KL-Divergence of a batch of distributions and a unit Gaussian.

Return type: Tensor

static configure()[source]

Return the default configuration for the BCE loss.

Return type: Dict

class ecgan.utils.losses.GANBaseLoss(discriminator_sampler, generator_sampler)[source]

Bases: ecgan.utils.configurable.Configurable

Base loss class for custom GAN losses.

abstract forward(training_data)[source]

Return the loss of a given component.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for a general loss function.

Return type: Dict

class ecgan.utils.losses.BceGeneratorLoss(discriminator_sampler, generator_sampler, reduction)[source]

Bases: ecgan.utils.losses.GANBaseLoss

BCE Loss using the PyTorch implementation.

forward(training_data)[source]

Calculate the BCE loss and update a given optimizer to minimize the loss.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for the BCE loss.

Return type: Dict

class ecgan.utils.losses.BceDiscriminatorLoss(discriminator_sampler, generator_sampler, reduction)[source]

Bases: ecgan.utils.losses.BceGeneratorLoss

Two component BCE Loss using the PyTorch implementation.

The class assumes that corresponding BaseSampler`s for each component are implemented. The fake data is sampled by the `sample method of the provided generator. The BCE loss is commonly used when optimizing the discriminator of a vanilla GAN.

forward(training_data)[source]

Calculate the two component BCE loss and update a given optimizer to minimize the loss.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for the two component BCE loss.

Return type: Dict

class ecgan.utils.losses.WassersteinDiscriminatorLoss(discriminator_sampler, generator_sampler, gradient_penalty_weight=None, clipping_bound=None)[source]

Bases: ecgan.utils.losses.GANBaseLoss

Wasserstein loss for the discriminator.

static apply_gp(input_tensor, target_tensor)[source]

GP penalty is applied outside the forward call during optimization.

Return type: Tensor

forward(training_data)[source]

Calculate the Wasserstein distance and minimize it using a given optimizer.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

get_gradient_penalty(real_data, generated_data)[source]

Based on https://github.com/EmilienDupont/wgan-gp/blob/master/training.py.

Return type: Tensor

static configure()[source]

Return the default configuration for the Wasserstein discriminator loss.

Return type: Dict

class ecgan.utils.losses.WassersteinGeneratorLoss(discriminator_sampler, generator_sampler)[source]

Bases: ecgan.utils.losses.GANBaseLoss

Wasserstein loss for the discriminator.

forward(training_data)[source]

Calculate the Wasserstein distance and minimize it using a given optimizer.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for the Wasserstein generator loss.

Return type: Dict

class ecgan.utils.losses.AEGANGeneratorLoss(discriminator_sampler, generator_sampler)[source]

Bases: ecgan.utils.losses.GANBaseLoss

Loss function for the auto-encoder based GANs.

forward(training_data)[source]

Perform the forward for the AEGAN generator loss.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for the AEGAN generator loss.

Return type: Dict

class ecgan.utils.losses.AEGANDiscriminatorLoss(discriminator_sampler, generator_sampler)[source]

Bases: ecgan.utils.losses.GANBaseLoss

Discriminator loss for a AEGAN module.

forward(training_data)[source]

Perform a forward pass for the loss.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for the AEGAN discriminator loss.

Return type: Dict

class ecgan.utils.losses.VAEGANGeneratorLoss(discriminator_sampler, generator_sampler, reconstruction_loss, distribution, kl_beta, device)[source]

Bases: ecgan.utils.losses.GANBaseLoss

Generator loss for the VAEGAN module.

forward(training_data)[source]

Perform a forward pass for the loss.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for the VAEGAN generator loss.

Return type: Dict

class ecgan.utils.losses.SupervisedLossFactory[source]

Bases: object

Meta module for creating correct loss functions.

class ecgan.utils.losses.GANLossFactory[source]

Bases: object

Meta module for creating correct GAN loss functions.

class ecgan.utils.losses.AutoEncoderLoss(autoencoder_sampler, use_mse)[source]

Bases: ecgan.utils.configurable.Configurable

Base loss class for custom GAN losses.

forward(training_data)[source]

Perform a forward pass for the loss.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for a general loss function.

Return type: Dict

class ecgan.utils.losses.VariationalAutoEncoderLoss(autoencoder_sampler, use_mse, kl_beta, distribution, device)[source]

Bases: ecgan.utils.configurable.Configurable

Base loss class for custom GAN losses.

forward(training_data)[source]

Perform a forward pass for the loss.

Return type: Tuple[Union[Tensor, List[Tuple[str, Tensor]]], List[Tuple[str, Any]]]

static configure()[source]

Return the default configuration for the VAEGAN generator loss.

Return type: Dict