sakura.sakuraAE.sakuraAE

class sakura.sakuraAE.sakuraAE(config_json_path, verbose=False, suppress_train=False, suppress_tensorboardX=False)

Bases: object

A comprehensive class for SAKURA pipeline

This class manages overall workflow of SAKURA includeing model initialization, training, testing, and model inference or external model merging based on the configuration and argument settings.

Parameters:

• config_json_path (str) – Path to the configuration JSON file, which contains all the necessary settings for the class

• verbose (bool, optional) – Whether to enable verbose console logging, defaults to False

• suppress_train (bool, optional) – Whether to suppress model training, only setup dataset and model, defaults to False

• suppress_tensorboardX (bool, optional) – Whether to suppress Logger to initiate tensorboardX (to prevent flushing logs), defaults to False

Methods

execute_inference	Perform inference on the given tasks represented as a list of stories.
generate_splits	Generate dataset split masks for model training and testing.
insert_external_module	Insert an external module and merge it with SAKURA model.
integrity_check	Perform integrity check on selected phenotypes/signatures against the input dataset.
load_checkpoint	Load a checkpoint file and resume the model's state, including parameters, random states, training progress, etc.
save_checkpoint	Save the current state of the model and training process as a checkpoint.
setup_dataset	Set up dataset for SAKURA model.
test	Test all latents of the model, with options to evaluate specific loss groups and dump selected latents.
train	Batch train model for at least one epoch.
train_hybrid	Train the model in hybrid mode, where model module splits are trained with flexibility.
train_hybrid_fastload	Implement the multithread dataloader version of hybrid mode training, where model module splits are trained with flexibility.
train_story	Train the model on the given sets of tasks represented as a list of storylines.