Train API

IMPROVE general training parameters

log_level: Type: str

Default: “DEBUG”

Help: Set log levels. Default is WARNING. Levels are: DEBUG, INFO, WARNING, ERROR, CRITICAL, NOTSET.
input_dir: Type: str

Default: “./”

Help: Base directory for input data. All additional input pathes will be relative to the base input directory.
output_dir: Type: str

Default: “./”

Help: Base directory for output data. All additional relative output pathes will be placed into the base output directory.
config_file: Type: str

Default: None

Help: Config file in INI format.
param_log_file: Type: str

Default: “param_log_file.txt”

Help: Log of final parameters used for run. Saved in output_dir if file name, can be an absolute path.
data_format: Type: str

Default: “.parquet”

Help: File format to save the ML data file (e.g., ‘.pt’, ‘.tfrecords’).
input_supp_data_dir: Type: str

Default: None

Help: Dir containing supplementary data in addition to benchmark data (usually model-specific data).
model_file_name: Type: str

Default: “model”

Help: Filename to store trained model (str is w/o file_format).
model_file_format: Type: str

Default: “.pt”

Help: File format to save the trained model.
epochs: Type: int

Default: 7

Help: Training epochs.
learning_rate: Type: float

Default: 7

Help: Learning rate for the optimizer.
batch_size: Type: int

Default: 7

Help: Training batch size.
val_batch: Type: int

Default: 64

Help: Validation batch size.
loss: Type: str

Default: “mse”

Help: Prediction performance metric to monitor for early stopping during model training (e.g., ‘mse’, ‘rmse’).
patience: Type: int

Default: 20

Help: Iterations to wait for a validation metric to get worse before stop training.
metric_type: Type: str

Default: “regression”

Help: Metrics appropriate for given task. Options are ‘regression’ or ‘classification’.

Drug Response Prediction training parameters

y_col_name: Type: str

Default: “auc”

Help: Column name in the y data file (e.g., response.tsv), that represents the target variable that the model predicts.

Synergy training parameters

y_col_name: Type: str

Default: “loewe”

Help: “Column name in the y data file (e.g., synergy.tsv), that represents the target variable that the model predicts.

In synergy prediction problem it can be one of [‘loewe’, ‘bliss’, ‘zip’, ‘hsa’, ‘smean’, ‘css’].”