cellflow.model.CellFlow.train

CellFlow.train(num_iterations, batch_size=1024, valid_freq=1000, callbacks=[], monitor_metrics=[], out_of_core_dataloading=False)

Train the model.

Note

A low value of 'valid_freq' results in long training because predictions are time-consuming compared to training steps.

Parameters:

• num_iterations (int) – Number of iterations to train the model.

• batch_size (int) – Batch size.

• valid_freq (int) – Frequency of validation.

• callbacks (Sequence[BaseCallback]) –

  Callbacks to perform at each validation step. There are two types of callbacks: - Callbacks for computations should inherit from

  ComputationCallback see e.g. cellflow.training.Metrics.

  • Callbacks for logging should inherit from LoggingCallback see e.g. WandbLogger.

• monitor_metrics (Sequence[str]) – Metrics to monitor.

• out_of_core_dataloading (bool) – If True, use out-of-core dataloading. Uses the cellflow.data._dataloader.OOCTrainSampler to load data that does not fit into GPU memory.

Return type:

None

Returns:

: Updates the following fields:

• cellflow.model.CellFlow.dataloader - the training dataloader.

• cellflow.model.CellFlow.solver - the trained solver.