Special

Special functions for neural networks.

nabla.ops.special.softmax(arg, axis=-1)

Compute softmax function in a numerically stable way.

Parameters:

• arg (Array) – Input array

• axis (int) – Axis along which to compute softmax

Returns:

Array containing softmax probabilities

Return type:

Array

nabla.ops.special.logsumexp(arg, axis=None, keep_dims=False)

Compute log(sum(exp(x))) in a numerically stable way.

Parameters:

• arg (Array) – Input array

• axis (int | None) – Axis along which to compute logsumexp. If None, compute over all elements.

• keep_dims (bool) – Whether to keep reduced dimensions

Returns:

Array containing logsumexp values

Return type:

Array

nabla.ops.special.where(condition, x, y)

Element-wise selection from x or y based on condition.

Parameters:

• condition (Array) – Boolean array for selection

• x (Array) – Array to select from where condition is True

• y (Array) – Array to select from where condition is False

Returns:

Array with elements selected from x or y

Return type:

Array

nabla.ops.special.cond(condition, true_fn, false_fn, *args, **kwargs)

Conditional execution based on a boolean condition.

Parameters:

• condition (Array) – Boolean array determining which function to execute

• true_fn (Callable) – Function to execute if condition is True

• false_fn (Callable) – Function to execute if condition is False

• *args – Arguments passed to the selected function

• **kwargs –

  Arguments passed to the selected function

Returns:

Result of the executed function

Return type:

Array