Calculate DZ Method

Functions

calculate_dZ(W_next, dZ_next, Z, activation)

Calculate dZ for hidden layers.

Parameters:

Name	Type	Description	Default
W_next	ndarray	Weights of the next layer.	required
dZ_next	ndarray	dZ of the next layer.	required
Z	ndarray	Linear output of the current layer.	required
activation	str	Activation function of the current layer.	required

Returns:

Type	Description
	numpy.ndarray: Gradient of Z.

Example

dZ = calculate_dZ(W_next, dZ_next, Z, 'relu')
print(dZ.shape)

Source code in microkeras/operations/backward/calculate_dZ.py

def calculate_dZ(W_next, dZ_next, Z, activation):
    """
    Calculate dZ for hidden layers.

    Args:
        W_next (numpy.ndarray): Weights of the next layer.
        dZ_next (numpy.ndarray): dZ of the next layer.
        Z (numpy.ndarray): Linear output of the current layer.
        activation (str): Activation function of the current layer.

    Returns:
        numpy.ndarray: Gradient of Z.

    Example:
        ```python
        dZ = calculate_dZ(W_next, dZ_next, Z, 'relu')
        print(dZ.shape)
        ```
    """
    activation_derivatives = {
        'sigmoid': sigmoid_derivative,
        'relu': relu_derivative,
        'linear': linear_derivative
    }
    activation_derivative = activation_derivatives[activation]
    return np.dot(W_next.T, dZ_next) * activation_derivative(Z)