search

idptools / parrot

Python package for protein sequence-based bidirectional recurrent neural network. Generalizable to a variety of protein bioinformatic applications.
MIT License
16 stars 2 forks source link

Parrot failing to find Cython #10

Closed jlotthammer closed 2 years ago

jlotthammer commented 2 years ago

It seems that we have an issue with Parrot finding cython. Here are the details I was able to gather.

Error message: warning in stationary: failed to import cython module: falling back to numpy warning in coregionalize: failed to import cython module: falling back to numpy warning in choleskies: failed to import cython module: falling back to numpy

Version information: Python 3.7.6 parrot.version = 'v1.6.2+0.gd17a873.dirty' Cython.version = 0.29.24

OS: Distributor ID: Ubuntu Description: Ubuntu 20.04.3 LTS Release: 20.04 Codename: focal

cc: @degriffith @alexholehouse we should try to look into these and resolve, I'd be interested to know if you guys also have the same issues - it appears that ryan does as well.

degriffith commented 2 years ago

I've seen this before, a while back. If I remember correctly this is a GPy version vs Python version issue. I will look into this more tomorrow.

jlotthammer commented 2 years ago

Ah darn GPy! Is it still actively maintained? I know we briefly chatted about this on slack because another incompatibility with torch, but I kinda figured we wouldn’t do it, but now I’m wondering if we should more seriously consider migrating to another hyper parameter tuner? This sorta depends on the complexity of the underlying code and the time investment to fix but I can open another issue if we’d like to discuss and consider (could very well be that we encounter other issues with something else, so I’m not trying to say let’s jump ship just yet, but worth considering our options). I know I liked the look of optuna and FWIW it’s what our friends at Microsoft were using for that GVP paper in the bowman lab, but another one that I’m aware of is hyperopt.

jlotthammer commented 2 years ago

FWIW in case there was any doubt, this is an issue with Gpy, because the detachment of parrot-optimize from the main parrot PyPI distribution does prohibit this numpy fallback from happening.