LeakyClamp

class pfhedge.nn.LeakyClamp(clamped_slope=0.01, inverted_output='mean')

Leakily clamp all elements in input into the range $[min,max]$.

The bounds $min$ and $max$ can be tensors.

If $min\le max$:

$$\begin{array}{r}\text{output}=\{\begin{array}{ll}min+\text{clampled\_slope}\ast (\text{input}-min)& \text{input}<min\\ \text{input}& min\le \text{input}\le max\\ max+\text{clampled\_slope}\ast (\text{input}-max)& max<\text{input}\end{array}\end{array}$$

If $min>max$:

$$\begin{array}{r}\text{output}=\{\begin{array}{ll}\frac{1}{2}(min+max)& \text{inverted\_output}=\text{'mean'}\\ max& \text{inverted\_output}=\text{'max'}\end{array}\end{array}$$

See also

• pfhedge.nn.functional.leaky_clamp()

Parameters

• clamped_slope (float, default=0.01) – Controls the slope in the clampled regions.

• inverted_output ({'mean', ''max'}, default='mean') – Controls the output when $min>max$. ‘max’ is consistent with torch.clamp().

Shape:

• input: $(N,\ast )$ where $\ast$ means any number of additional dimensions.

• min: $(N,\ast )$ or any size broadcastable to input.

• max: $(N,\ast )$ or any size broadcastable to input.

• output: $(N,\ast )$, same shape as the input.

Examples

>>> import torch
>>> from pfhedge.nn import LeakyClamp
>>> m = LeakyClamp()
>>> input = torch.linspace(-2, 12, 15) * 0.1
>>> input
tensor([-0.2000, -0.1000,  0.0000,  0.1000,  0.2000,  0.3000,  0.4000,  0.5000,
         0.6000,  0.7000,  0.8000,  0.9000,  1.0000,  1.1000,  1.2000])
>>> m(input, 0.0, 1.0)
tensor([-2.0000e-03, -1.0000e-03,  0.0000e+00,  1.0000e-01,  2.0000e-01,
         3.0000e-01,  4.0000e-01,  5.0000e-01,  6.0000e-01,  7.0000e-01,
         8.0000e-01,  9.0000e-01,  1.0000e+00,  1.0010e+00,  1.0020e+00])

forward(input, min=None, max=None)

Clamp all elements in input into the range $[min,max]$.

Parameters

• input (torch.Tensor) – The input tensor.

• min (torch.Tensor, optional) – Lower-bound of the range to be clamped to.

• max (torch.Tensor, optional) – Upper-bound of the range to be clamped to.

Returns

torch.Tensor