scripts package

Submodules

scripts.run_ar_tc_ignite module

Example script for running DeepSphere U-Net on reduced AR_TC dataset.

scripts.run_ar_tc_ignite.add_tensorboard(engine_train, optimizer, model, log_dir)

Creates an ignite logger object and adds training elements such as weight and gradient histograms

Parameters:

• engine_train (ignite.engine) – the train engine to attach to the logger
• optimizer (torch.optim) – the model’s optimizer
• model (torch.nn.Module) – the model being trained
• log_dir (string) – path to where tensorboard data should be saved

scripts.run_ar_tc_ignite.average_precision_compute_fn(y_pred, y_true)

Attached function to the custom ignite metric AveragePrecisionMultiLabel

Parameters:

• y_pred (torch.Tensor) – model predictions
• y_true (torch.Tensor) – ground truths

Raises:

RuntimeError – Indicates that sklearn should be installed by the user.

Returns:

average precision vector.

Of the same length as the number of labels present in the data

Return type:

numpy.array

scripts.run_ar_tc_ignite.get_dataloaders(parser_args)

Creates the datasets and the corresponding dataloaders

Parameters:parser_args (dict) – parsed arguments Returns:train, validation dataloaders Return type:(torch.utils.data.dataloader, torch.utils.data.dataloader)

scripts.run_ar_tc_ignite.main(parser_args)

Main function to create trainer engine, add handlers to train and validation engines. Then runs train engine to perform training and validation.

Parameters:parser_args (dict) – parsed arguments

scripts.run_ar_tc_ignite.validate_output_transform(x, y, y_pred)

A transform to format the output of the supervised evaluator before calculating the metric

Parameters:

• x (torch.Tensor) – the input to the model
• y (torch.Tensor) – the output of the model
• y_pred (torch.Tensor) – the ground truth labels

Returns:

model predictions and ground truths reformatted

Return type:

(torch.Tensor, torch.Tensor)

Module contents

DeepSphere Base Documentation doc