Failed to reproduce the PytorchModel examples

[nychiang]

It seems that the function PytorchModel is not implemented (see here).
Are there plans to implement it? Or how we can reproduce the example?

@cnpetra
@prata74

[odow]

Ah, the docs are missing the instruction

import PythonCall

The code is here:
https://github.com/lanl-ansi/MathOptAI.jl/blob/main/ext/MathOptAIPythonCallExt.jl

It gets loaded in the CI tests because I have

MathOptAI.jl/test/test_PythonCall.jl

Line 15 in 42017ae

import PythonCall

and it gets loaded in the documentation build because I have

MathOptAI.jl/docs/make.jl

Line 16 in 42017ae

import PythonCall

I need to clarify this in the docs. Or perhaps I should just add PythonCall as a strict dependency.

Thoughts @Robbybp? I initially wanted to keep the deps as light as possible, but it seems like PythonCall might be a pretty common use case (much more than say, Lux or Flux).

[odow]

I just made a PR to improve the documentation: #163

[Robbybp]

I say keep the dependencies light and clarify the docs

[odow]

As an audit: the other extensions are okay, because users explicitly need to import them to build the ML (you can't load a Flux model without using Flux). PythonCall is the only exception because we need PythonCall, but the user doesn't need it.

[odow]

Thanks for opening the issue @nychiang. Don't hesitate if you have other questions 😄

[odow]

Release docs are now updated:
https://lanl-ansi.github.io/MathOptAI.jl/stable/api/#PytorchModel

https://lanl-ansi.github.io/MathOptAI.jl/stable/tutorials/pytorch/