Ep/qmml #666
Ep/qmml #666
Conversation
The molecular charge and molecular spin embedding taken from SpookyNet and slighly modified for SchNet representation.
testing the electronic_embedding module databases for carbene and ag3 ions train_V01.py is the main file to run the training train.sh is the bash script to run the training on the cluster
add electronic embedding to so3 net and bugfix painn and schnet rep
… all. Plus callback to set lr on value if specific validation loss threshold reached
… confirmed with carbon that requested atom type is filtered out
…is kind of hardcoded and oriented for sparse representation of transition metals in qmml dataset, and should be revisited later. But for now the first two to be investigated cases carbon exclusion and TM exclusion can be done
.vscode/extensions.json
Outdated
There was a problem hiding this comment.
I think this should be ignored
| @@ -0,0 +1 @@ | |||
| include src/schnetpack/train/ressources/partition_spline_for_robust_loss.npz | |||
There was a problem hiding this comment.
From this file values are loaded which are used to approximate the partition function to for the adaptive loss fn
There was a problem hiding this comment.
I would keep this private/local and not upload to schnetpack
There was a problem hiding this comment.
Do you mean because of the data ? or the attached comments ?
| batch_size: 50 | ||
| num_train: 0.90 | ||
| num_val: 0.05 | ||
| load_properties: [formation_energy,forces,charge,multiplicity] |
There was a problem hiding this comment.
I don`t think it is possible to pass a list like this. If I remember correctly it should work like
load properties:
- formation_energy
- forces
- ...
There was a problem hiding this comment.
It is possible to pass a list like this.
| @@ -57,5 +57,8 @@ script-files = [ | |||
| "src/scripts/spkdeploy", | |||
| ] | |||
|
|
|||
| # Ensure package data such as resources are included | |||
| package-data = { "schnetpack.train" = ["ressources/partition_spline_for_robust_loss.npz"] } | |||
There was a problem hiding this comment.
this is needed to include the file when building the package
There was a problem hiding this comment.
I would maybe directly add this in the experiment yaml file and not as a separate config file, because it is barely used at the moment
| @@ -1,5 +1,6 @@ | |||
| defaults: | |||
| - radial_basis: gaussian | |||
| - nuclear_embedding: complex | |||
There was a problem hiding this comment.
remove this and leave default value as it is
There was a problem hiding this comment.
I agree, done in 0f6fc8b
| scheduler_cls: torch.optim.lr_scheduler.MultiStepLR | ||
| scheduler_monitor: val_loss | ||
| scheduler_args: | ||
| milestones: [3,9,15,18,24,30,36] |
There was a problem hiding this comment.
the milestones are very specific to the experiment. I would either set them to ??? or just define the scheduler in the experiment.yaml
There was a problem hiding this comment.
I agree, done in ca4acd7
src/schnetpack/data/atoms.py
Outdated
| @@ -344,8 +344,10 @@ def _get_properties( | |||
| properties = {} | |||
| properties[structure.idx] = torch.tensor([idx]) | |||
| for pname in load_properties: | |||
|
|
|||
There was a problem hiding this comment.
remove blank lines
Contains changes to work with QCML Dataset