[JOSS Review] JOSS Paper Comments

[timothyas]

Hi there, really great work on this package and JOSS paper. The paper is very well written and well motivated, I just have two comments for you.

1. I know the focus of this work is on pytorch rather than Keras, but I think that given the overlap in content, it would be worth mentioning the Fortran Keras Bridge (here's the github repo and arxiv paper).
2. Tiny typo: line 102 should say "trained" not "trainied".

Otherwise the paper looks great!

cc: openjournals/joss-reviews#7602 @matthewfeickert

[jatkinson1000]

Hi @timothyas thanks for your helpful feedback!

Thanks for raising this.
We discussed this during writing and decided that we would not include FKB as it feels that it has moved towards abandonware - no updates in over 4 years, and unmerged PRs asking to upgrade to be compatible with keras 2.
The final decision was made when we saw that neural-fortran, which we do cite, states on the repo that

"As of v0.9.0, neural-fortran implements the full feature set of FKB in pure Fortran, and in addition supports training and inference of convolutional networks."

However, if you still feel it should be included we can do so.
We could also perhaps include fortran-tf-lib as a similar library that works with more recent tensorflow/keras.

Let me know your thoughts.

Typo I'll look at correcting shortly :)

[jatkinson1000]

I have added a commit in #200 that fixes the typo.

[timothyas]

Thanks @jatkinson1000, that is good to know (and TIL the term Abandonware... nice). Since FKB seems to be history, I don't think it's necessary to mention in the paper.

That said, I think it would be good to mention the fortran-tf-lib as a similar library just for people's awareness.