#450 Package psyneulink: A block modeling system for cognitive neuroscience
Opened 2 years ago by ankursinha. Modified 2 years ago

Please use this ticket to add new software to the NeuroFedora packaging queue.

New software: psyneulink

Short description: A block modeling system for cognitive neuroscience

Upstream URL: https://princetonuniversity.github.io/PsyNeuLink/

License: ASL

Domain: Computational modelling

Additional information

I started taking a look at the dependencies.

Upstream puts a maximum version on everything, and those limitations will have to be removed across the board.

Main package

There is a hard dependency on PyTorch in requirements.txt on platforms for which it exists, but it is actually optional (the program falls back gracefully), so this, too, can be patched out.

What’s left that isn’t in Fedora is:

Documentation

Probably not worth it to try to package the documentation.

If it is packaged, it will have to be patched to use a different theme, since the custom theme (https://github.com/dillontsmith/psyneulink_sphinx_theme), forked from https://github.com/pytorch/pytorch_sphinx_theme, is too much trouble at best: it compiles CSS from SCSS, which we would have to replicate in an RPM build; and it bundles several fonts, which we would have to package separately, patching the theme to use unbundled versions in system instead of web font formats.

tutorial Python extra

Sure, this just brings in graphviz, jupyter, and matplotlib.

cuda Python extra

Obviously, this must be omitted.

I think llvmlite is on our list for numba too, but there were issues with it when I looked at it last. I think it didn't support the latest llvm back then, but maybe it does so now:

https://github.com/numba/llvmlite

If it works fine without Pytorch but uses Pytorch if it is installed, that's fine with us. Folks can install Pytorch themselves if they need it.

Actually, https://github.com/dillontsmith/leabra has a new PyPI name compared to https://github.com/benureau/leabra, but the import is still leabra in both cases. The README in the fork explains that it was made only because the original authors did not upload to PyPI. In Fedora, it may be just as well to package the original instead.

Login to comment on this ticket.

Metadata
Boards 1
Software packaging Status: Backlog