Testing hpo packages #218
Testing hpo packages #218
Conversation
|
Can we save this qm9 as a separate directory (say, qm9_hpo)? We often use qm9 as a minimum case. It would be great if we make "qm9" as simple as possible. How do you think? |
@jychoi-hpc agree. |
allaffa
force-pushed
the
testing_hpo_packages
branch
from
e98daff to
d814164
Compare
allaffa
force-pushed
the
testing_hpo_packages
branch
from
d814164 to
3c75e5e
Compare
There was a problem hiding this comment.
The PR looks good to me overall. I'll need to learn more about these hpo tools. Three general comments: 1) remove or clean up unused functions in dh_utils.py; 2) save trained models in all trials in qm9.py; 3) maybe add data loaders as command line arguments to avoid repeated data preprocessing in trails.
| config["NeuralNetwork"], | ||
| log_name, | ||
| verbosity, | ||
| ) |
There was a problem hiding this comment.
save_model after training
There was a problem hiding this comment.
Thanks. I added these lines.
|
@pzhanggit |
* qm9_hpo folder added as separate test * pei comments addressed * num_headlayers and dim_headlayers added to hyperparameter list * black formatting fixed
This PR introduces three main examples for hyperparameter optimization (HPO) using the QM9 dataset:
qm9_optuna.py: HPO example usingoptunaqm9_deephyper.py: HPO example running in sequential usingdeephyperqm9_deephyper_multi.py: HPO example running in parallel usingdeephyper