Handle multiple outputs from models.

jatkinson1000 commented 3 months ago

Currently work in progress.

Requested by user in #135 Closes #3

Definitely possible, but my C++-fu is not as strong as Fortran so currently stuck in pointer hell.

[x] Update C++ to process the option of multiple outputs
- [x] Inspect result to see if it is a tensor or tuple and handle accordingly
[x] Update Fortran to handle this
- [x] Make outputs a vector like inputs
- [x] Require n_out as argument to forward

Other thoughts:

This will be a breaking API change for current users 😞
[x] Should we make this an interface so users can pass a single tensor or an array of tensors?
[x] Can we somehow detect n_in? I have a feeling we looked before and the answer was no...
- The answer is, in fact, yes - see https://github.com/Cambridge-ICCS/FTorch/pull/136/commits/4ab920a5d9e0260a30b3c844aeeac06223910fbf

jwallwork23 commented 3 months ago

Okay @jatkinson1000 @TomMelt @ma595 this is now working! Ready for review.

jwallwork23 commented 3 months ago

Updated docs - ready for re-review.

jatkinson1000 commented 3 months ago

@TomMelt It seems like this was merged somewhat prematurely. As a result I am having to update the README over in #143 to cover some of the changes made here which is not ideal.

Cambridge-ICCS / FTorch

Handle multiple outputs from models. #136