constraints

NonNegative

Bases: tf.keras.constraints.Constraint

Constraint that enforces positive activations.

__call__

__call__(w: tf.Tensor) -> tf.Tensor

Clips negative values of a provided weight tensor w to 0.0 and leaves positive values unchanged.

Parameters:

Name	Type	Description	Default
w	tf.Tensor	Tensor containing the wehights of a layer.	required

Returns:

Type	Description
tf.Tensor	Tensor where negative values are clipped to 0.0.

Source code in DeepSaki/constraints/constraints.py

def __call__(self, w: tf.Tensor) -> tf.Tensor:
    """Clips negative values of a provided weight tensor `w` to 0.0 and leaves positive values unchanged.

    Args:
        w (tf.Tensor): Tensor containing the wehights of a layer.

    Returns:
        Tensor where negative values are clipped to 0.0.
    """
    return w * tf.cast(tf.math.greater_equal(w, 0.0), w.dtype)