Test GraphDRP

[wilke]

Command: train.sh 4 ./tmp Status:

[wilke]

Commands:

1. git clone git@github.com:JDACS4C-IMPROVE/GraphDRP.git
2. conda create -n GraphDRP python=3.7 pip --yes
3. conda activate GraphDRP
4. conda install pytorch torchvision cudatoolkit=10.2 -c pytorch --yes
5. conda install pyg -c pyg -c conda-forge --yes
6. conda install -c conda-forge matplotlib --yes
7. conda install -c conda-forge h5py=3.1 --yes
8. conda install -c bioconda pubchempy --yes
9. conda install -c rdkit rdkit --yes
10. conda install -c anaconda networkx --yes
11. conda install -c conda-forge pyarrow=10.0 --yes
12. conda install -c pyston psutil --yes
13. pip install git+https://github.com/ECP-CANDLE/candle_lib@develop
14. cd GraphDRP && mkdir -p tmp
15. train.sh 4 ./tmp

Output:

...
MSE improved at epoch 2; Best MSE: 0.00179358; Model: GINConvNet; Dataset: GDSC
Make prediction for 19105 samples...
Runtime: 2.8 mins
Scores:
        {'total_epochs': 2, 'best_epoch': 2, 'val_loss': 0.0017935786163434386, 'pcc': 0.8193430798183505, 'scc': 0.7715892898977942, 'rmse': 0.04235066252543682, 'r2': 0.601891340527829}

IMPROVE_RESULT val_loss:        0.0017935786163434386

Finished.

[wilke]

Status: OK

[wilke]

Testing also singularity exec --bind ${TEST_DIR}:/candle_data_dir build/GraphDRP.sif train.sh 2 /candle_data_dir

[wilke]

Output:

Train epoch: 2 [3895680/152839 (80%)]   Loss: 0.002414
Make prediction for 19105 samples...
MSE improved at epoch 2; Best MSE: 0.00179435; Model: GINConvNet; Dataset: GDSC
Make prediction for 19105 samples...
Runtime: 3.3 mins
Scores:
        {'total_epochs': 2, 'best_epoch': 2, 'val_loss': 0.0017943511484190822, 'pcc': 0.8144680153400706, 'scc': 0.7711946625318125, 'rmse': 0.042359782204575626, 'r2': 0.6017198716245734}

IMPROVE_RESULT val_loss:        0.0017943511484190822

Finished.

Status: OK