Please have a look at the Chemprop documentation for details about these commands.
The data used are:
- For model training and validation:
data/article_training.csv - For predictions:
data/coco_test.csv
Training using Chemprop's default settings, and data/article_training.csv as our data:
chemprop_train --data_path data/article_training.csv --dataset_type classification --save_dir article_checkpoints
One first needs to make sure that there is a file available for the optimization script to save its results to.
The number of different models explored is sepecified through --num_iters.
# Create empty file for storing the hyperopt results
touch hyperopt_configs.json
# Run hyperopt
chemprop_hyperopt --data_path data/article_training.csv --dataset_type classification --num_iters 20 --config_save_path hyperopt_configs.json
The checkpointed model used is the same as the one from the training phase, but the data is now different: data/coco_test.csv.
chemprop_predict --test_path data/coco_test.csv --checkpoint_dir article_checkpoints --preds_path coco_preds.csv
Model interprability: Understanding which chemical groups might be responsible for various properties.
chemprop_interpret --data_path <path> --checkpoint_dir <dir> --property_id 1