search

lanl-ansi / MathOptAI.jl

Embed trained machine learning predictors in JuMP
https://lanl-ansi.github.io/MathOptAI.jl/
Other
29 stars 1 forks source link

Raise error if Torch Softplus uses a custom beta #112

Closed Robbybp closed 3 weeks ago

Robbybp commented 4 weeks ago

Torch's softplus is parameterized by a number, beta, that (I think) controls how closely the function approximates ReLU. I propose we raise an error if this parameter is set (to anything other than 1). The other option would be to support this in our SoftPlus implementation. From a quick scan, Keras and Flux don't have this parameter, so I think we probably don't need to support it either.

odow commented 4 weeks ago

I don't think it's that hard to support? It seems reasonable.