search

deepchem / moleculenet

Moleculenet.ai Datasets And Splits
MIT License
88 stars 19 forks source link

Create MolNet_Benchmark_Tutorial.ipynb #39

Closed ncfrey closed 3 years ago

ncfrey commented 3 years ago

I wrote a brief tutorial that shows how to run a MolNet benchmark in Google Colab. This was helpful to me to set up a consistent environment and run on the GPU.

If it would be helpful, I'd like to add the tutorial to the repo. Any feedback is welcome!

CCing @mufeili @rbharath

mufeili commented 3 years ago

Can we replace

!conda install -c dglteam dgl-cuda11.0 dgllife

with

!conda install -c dglteam dgl-cuda11.0
!pip install dgllife

I deprecated the conda distributions for DGL-LifeSci since 0.2.7.

mufeili commented 3 years ago

Can you explain what is /content?

mufeili commented 3 years ago

"hyperparam" -> "hyperparameter"

mufeili commented 3 years ago

"The hyperparam search space can be modified in the molnet python script if desired." Can you specify the exact script name(s)?

mufeili commented 3 years ago

For "16 different models", it's not clear what are the 16 models as only RF and GraphConv were introduced. It's probably better to say "16 RF models with different hyperparameters".

mufeili commented 3 years ago

I'm good otherwise. Great job!

ncfrey commented 3 years ago

Can you explain what is /content?

This was left in unnecessarily, I've removed it.

For "16 different models", it's not clear what are the 16 models as only RF and GraphConv were introduced. It's probably better to say "16 RF models with different hyperparameters".

Thanks for pointing this out, I've changed the text to read: After training the random forest model with 16 different hyperparameter configurations (and repeating each 3 times to collect statistics for the validation and test errors)

I've updated the tutorial to address @mufeili's comments. Thanks for the feedback!