Closed rishigurnani closed 1 year ago
I just tested the same dataset on main and the new branch. The one on main ran, but the one in the new branch errored.
Traceback pasted below
..Training model 0 (with capacity 2) of 4
Traceback (most recent call last):
File "GNN_CV_training.py", line 362, in
One change is that an argument was added to the
polyGNN
class to specify the dimension ofgraph_feats
.
Did you specify the dimension of graph_feats
using the new argument graph_feats_dim
?
Actually I realized there's another issue. I'll let you know when it's fixed.
@oliverhvidsten OK, now please try again. You'll need the new polygnn_trainer
version (v0.5.0) as well. Do you have it? If not, please run poetry update polygnn_trainer
.
Getting this error. Any thoughts?
Traceback (most recent call last):
File "GNN_CV_training.py", line 369, in
It's not obvious to me why that is happening, since example2.py
runs without issue. Can you send a reproducible example?
Oops. Forgot to pull new changes. That error is no longer occurring.
The plots are no longer invariant across the features contained within graph_feats. The accuracy of the predictions I just made were not great, but it is likely because I was using a very small part of the overall dataset. I will test again with the full dataset and see what information I get back. This will take a bit though.
The plots are no longer invariant across the features contained within graph_feats. The accuracy of the predictions I just made were not great, but it is likely because I was using a very small part of the overall dataset. I will test again with the full dataset and see what information I get back. This will take a bit though.
Awesome! I'll go ahead and close #13 then. When you get the parity plots generated using the new code and the full dataset can you do me a favor and add them to #13 for posterity? You can reopen the issue too then if necessary.
Prior to this PR,
graph_feats
was not being used in the forward pass. This meant that predictions would be the same for polymer with the same SMILES and selector, even if they had different graph_feats. This issue is fixed in this PR. One change is that an argument was added to thepolyGNN
class to specify the dimension ofgraph_feats
.I also added a test to make sure in the future that this bug does not occur again.