-
Notifications
You must be signed in to change notification settings - Fork 548
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Backwards compatibility for Missing Training Args #149
Conversation
This pull request introduces 2 alerts when merging d874a0f into c60973b - view on LGTM.com new alerts:
|
Have tested this with multiple molecules and with molecule features. |
d874a0f
to
357a09e
Compare
I think this is great (especially to have all the checks in a separate functions, since this also cleans up large portions in |
@hesther I think this is an important concern and it's one that I was trying to think about as well. It's one of the reasons I was looking so close at continuity yesterday. My only problem was that I couldn't find an example where the default wasn't the old behavior. I was using only fairly basic arguments in the search so it makes sense that I overlooked the atom features scaling. My solution to this would be (now and in the future) to make a hard override list in the function to use instead of defaults if the default behavior is not the previous behavior. This should be a rare occurrence. I'll go and add the feature scaling argument to such a list in the function. If we can identify any others like this, we should add them as well. |
The atom and bond features scaling (original args and intermediate value args) are included now in the override list. It occurs to me that necessary intermediate value args may not be caught in the current structure, if the argument is not set during |
31035dc
to
67b42b6
Compare
@mliu49 this fix should be ready to merge |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I added a bunch of minor comments and suggestions. I think the code consolidation is great!
@mliu49 thanks again for all the good notes! They should all be implemented now. |
Add required dataset_type for default args Remove Unused Code override defaults for atom and bond features scaling Finish adding the override argument mechanism Updates for backwards compatibility Simplify default override list
fa0000b
to
50b4740
Compare
Fix for #148. This PR adds a function that fills in the default values for training args when those training args are missing in the checkpoint file. This bridges a backwards compatibility issue introduced at #137.