search

alan-turing-institute / grace

Graph Representation Analysis for Connected Embeddings
Other
34 stars 1 forks source link

New conda environments don't train from `run.py` #159

Open KristinaUlicna opened 1 year ago

KristinaUlicna commented 1 year ago

Neither of the new conda envs proceeds to train the GCN model from run.py script. Error:

zsh: segmentation fault python3 grace/training/run.py

This applies to both grace-env-with-napari and grace-env-napari-free

Not sure what's going on, the training works with other environments. @mooniean could you see if training works for you? 🥲

mooniean commented 1 year ago

The segmentation fault appears to happen when using import torch.

In manually installed environments (including the one from where the yaml was created), this error is not reproduced, however in packages installed from conda create, it does happen. From the solution here, the segmentation fault doesn't happen if we add the import numpy before import torch.