This document provides instructions on how to use the polymon command-line interface (CLI).

The polymon CLI has three main modes:

• Train: Train a machine learning or deep learning model.

• Merge: Merge two datasets.

• Predict: Predict labels for a given dataset.

Train

This command is used to train a model.

Usage:

polymon train [OPTIONS]

Arguments:

Supported Arguments in train	Type	Default	Description
--raw-csv	str	database/database.csv	Path to the raw csv file.
--sources	str (multiple)	['Kaggle']	Sources to use for training.
--tag	str	debug	Tag to use for training.
--labels	str (multiple)	Required	Labels to use for training.
--feature-names	str (multiple)	['rdkit2d']	Feature names to use for training.
--n-trials	int	None	Number of trials to run for hyperparameter optimization.
--out-dir	str	./results	Path to the output directory.
--hparams-from	str	None	Path to the hparams file. Allowed formats: .json, .pt, .pkl.
--n-fold	int	1	Number of folds to use for cross-validation.
--split-mode	str	random	Mode to split the data into training, validation, and test sets.
--seed	int	42	Seed to use for training.
--remove-hydrogens	bool	False	Whether to remove hydrogens from the molecules.
--descriptors	str (multiple)	None	Descriptors to use for training. For ML models, this must be specified.
--model	str	rf	Model to use for training.
--hidden-dim	int	32	Hidden dimension of the model.
--num-layers	int	3	Number of layers of the model.
--batch-size	int	128	Batch size to use for training.
--lr	float	1e-3	Learning rate to use for training.
--num-epochs	int	2500	Number of epochs to use for training.
--early-stopping-patience	int	250	Number of epochs to wait before early stopping.
--device	str	cuda	Device to use for training.
--run-production	bool	False	Whether to run the training in production mode, which means train:val:test splits will be forced to 0.95:0.05:0.0.
--finetune	bool	False	Whether to finetune the model.
--finetune-csv-path	str	None	Path to the csv file to finetune the model on.
--pretrained-model	str	None	Path to the pretrained model.
--n-estimator	int	1	Number of estimators to use for training.
--additional-features	str (multiple)	None	Additional features to use for training.
--skip-train	bool	False	Whether to skip the training step.
--low-fidelity-model	str	None	Path to the low fidelity model.
--estimator-name	str	None	Name of the estimator to give base predictions.
--emb-model	str	None	Name of the embedding model for base graph embeddings.
--ensemble-type	str	voting	Type of ensemble to use for training.
--train-residual	bool	False	Whether to train the residual of the model.
--normalizer-type	str	normalizer	Type of normalizer to use for training. Choices: normalizer, log_normalizer, none.
--augmentation	bool	False	Whether to use data augmentation.

Merge

This command is used to merge two datasets.

Usage:

polymon merge [OPTIONS]

Arguments:

Supported Arguments in merge	Type	Default	Description
--sources	str (multiple)	Required	Sources to merge.
--label	str	Required	Label to merge.
--hparams-from	str	Required	Path to the hparams file.
--acquisition	str	Required	Acquisition function to use for merging. Choices: epig, uncertainty, difference.
--sample-size	int	20	Sample size to use for merging.
--uncertainty-threshold	float	0.1	Uncertainty threshold to use for merging.
--difference-threshold	float	0.1	Difference threshold to use for merging.
--target-size	int	1000	Target size to use for merging.
--base-csv	str	None	Path to the base csv file.

Predict

This command is used to predict labels for a given dataset.

Usage:

polymon predict [OPTIONS]

Arguments:

Argument	Type	Default	Description
--model-path	str	Required	Path to the model.
--csv-path	str	Required	Path to the csv file.
--smiles-column	str	Required	Name of the smiles column.