pyml.neural_network.loss.categorical_cross_entropy.CategoricalCrossentropy#

class CategoricalCrossentropy[source]#

Bases: _Loss

Categorical Cross-Entropy loss function for multi-class classification tasks.

This class defines the forward and backward pass computations for calculating the Categorical Cross-Entropy loss and its gradient with respect to the predicted values.

Variables:: dinputs (numpy.ndarray) – Gradient of the loss with respect to the predicted values.

Methods

`__init__`
`backward`	Compute the backward pass to calculate the gradient of the loss with respect to the predicted values.
`calculate`	Calculate the total loss based on model output and ground truth values.
`calculate_accumulated`	Calculate the accumulated loss over a training pass.
`forward`	Compute the forward pass of the Categorical Cross-Entropy loss.
`regularization_loss`	Calculate the regularization loss based on L1 and L2 regularization terms of the trainable layers' weights and biases.
`reset`	Reset the accumulated loss variables for a new training pass.
`set_trainable_layers`	Set the list of trainable layers for loss calculations.

backward(dvalues, y_true)[source]#

Compute the backward pass to calculate the gradient of the loss with respect to the predicted values.

Return type:

None

Parameters:

dvalues (numpy.ndarray) – Gradient of the loss with respect to the predicted values.
y_true (numpy.ndarray) – Ground truth values, containing integer class labels or one-hot encoded vectors.

calculate(output, y, *, include_regularization=False)#

Calculate the total loss based on model output and ground truth values.

Return type:

ndarray | tuple[ndarray, float]

Parameters:

output (numpy.ndarray) – Predicted output from the model.
y (numpy.ndarray) – Ground truth values.
include_regularization (bool, optional) – Whether to include regularization loss, by default False.

Returns:

Total loss value if include_regularization is False, or a tuple of data and regularization loss values if include_regularization is True.

Return type:

numpy.ndarray or tuple

calculate_accumulated(*, include_regularization=False)#

Calculate the accumulated loss over a training pass.

Return type:: ndarray | tuple[ndarray, float]
Parameters:: include_regularization (bool, optional) – Whether to include regularization loss, by default False
Returns:: Accumulated loss value or tuple of data and regularization losses.
Return type:: float or tuple

forward(y_pred, y_true)[source]#

Compute the forward pass of the Categorical Cross-Entropy loss.

Return type:

ndarray

Parameters:

y_pred (numpy.ndarray) – Predicted output from the model, representing class probabilities.
y_true (numpy.ndarray) – Ground truth values, containing integer class labels or one-hot encoded vectors.

Returns:

Array of negative log-likelihoods for each sample.

Return type:

numpy.ndarray

regularization_loss()#

Calculate the regularization loss based on L1 and L2 regularization terms of the trainable layers’ weights and biases.

Return type:: float
Returns:: Regularization loss value.
Return type:: float

reset()#

Reset the accumulated loss variables for a new training pass.

Return type:: None

set_trainable_layers(trainable_layers)#

Set the list of trainable layers for loss calculations.

Return type:: None
Parameters:: trainable_layers (list[_Layer]) – List of trainable layers in the model.