Documentation for soft_indicator

Functionality

The soft_indicator function computes a differentiable indicator that approximates the behavior of a hard indicator (x < threshold). It utilizes a sigmoid function to smoothly transition between outputs as the input values approach the specified threshold.

Parameters

• x: A torch.FloatTensor containing input values.
• threshold: A float that defines the threshold, near which the indicator is approximately 0.5.
• steepness: An int controlling the sharpness of the transition.

Usage

• Purpose: To generate weights ranging from 0.0 to 1.0 based on a smooth comparison with a threshold.

Example

For example:

x = torch.tensor([0.005, 0.02])
soft_indicator(x, threshold=0.01, steepness=100)

might return a tensor similar to [1.0, 0.0].