| nnf_softplus {torch} | R Documentation |
Softplus
Description
Applies element-wise, the function Softplus(x) = 1/\beta * log(1 + exp(\beta * x)).
Usage
nnf_softplus(input, beta = 1, threshold = 20)
Arguments
input |
(N,*) tensor, where * means, any number of additional dimensions |
beta |
the beta value for the Softplus formulation. Default: 1 |
threshold |
values above this revert to a linear function. Default: 20 |
Details
For numerical stability the implementation reverts to the linear function
when input * \beta > threshold.
[Package torch version 0.13.0 Index]