bartorch.soft_thresh

bartorch.soft_thresh()

Soft thresholding, x * max(1 - lamda / |x|, 0).

Parameters:

• lamda (float) – Threshold, in the units of input’s magnitude.

• input (torch.Tensor)

• joint_axes (int or tuple of int, default=()) – Axes over which |x| is the Euclidean norm, so that each fibre along them is shrunk as one vector; empty for element-wise.

Returns:

The shape of input.

Return type:

torch.Tensor