yamle.losses.regression module

class yamle.losses.regression.GaussianNegativeLogLikelihoodLoss(flatten=False, *args, **kwargs)

Bases: BaseLoss

This defines the base negative log-likelihood loss.

It assumes that the input shape is (batch_size, num_members, 1) of only mean is predicted or (batch_size, num_members, 2) if mean and variance are predicted. The first feature is the mean and the second is the variance. No matter what the reduction it is always averaged over the num_members.

The loss can also be weighted by a weight tensor of shape (batch_size).

Parameters:

flatten (bool) – Whether to flatten the input. Defaults to False.

tasks = ['regression', 'depth_estimation', 'reconstruction']

static add_specific_args(parent_parser)

This method is used to add the specific arguments for this loss.

Return type:

ArgumentParser

class yamle.losses.regression.MeanSquaredError(flatten=False, *args, **kwargs)

Bases: BaseLoss

This defines the mean squared error loss.

It assumes that the input shape is (batch_size, num_members, 1). If there are more features, only the first feature is used. No matter what the reduction it is always averaged over the num_members.

The loss can also be weighted by a weight tensor of shape (batch_size).

Parameters:

flatten (bool) – Whether to flatten the input. Defaults to False.

tasks = ['regression', 'depth_estimation', 'reconstruction']

static add_specific_args(parent_parser)

This method is used to add the specific arguments for this loss.

Return type:

ArgumentParser

class yamle.losses.regression.QuantileRegressionLoss(quantiles=[0.1, 0.5, 0.9], **kwargs)

Bases: BaseLoss

This defines the quantile regression loss.

It assumes that the input shape is (batch_size, num_members, quantiles) The losses for different quantiles are averaged.

The loss can also be weighted by a weight tensor of shape (batch_size).

Parameters:

quantiles (List[float]) – The quantiles to be used for computing the loss.

tasks = ['regression', 'depth_estimation', 'reconstruction']

static add_specific_args(parent_parser)

This method is used to add the specific arguments for this loss.

Return type:

ArgumentParser