GLITask

class gli.task.GLITask(task_dict, pwd, device='cpu')

Bases: object

The basic GLI task class for creating graph learning tasks.

This class contains the necessary attributes and methods for all GLI tasks.

Parameters:

• task_dict (dict) – A dictionary containing the task information.

• pwd (str) – The path to the directory containing the task files.

• device (str) – The device to load the task data to.

Notes

The task_dict should at least contain the following keys:

• type: The type of the task.

• description: A description of the task.

• feature: A list of features to use for the task.

• target: The target to use for the task.

The gli.task.GLITask class also contains split information. There are two split methods supported by GLI tasks: random split and predefined split.

If the random split is used, the task_dict should contain the following keys:

• train_ratio: The ratio of training samples.

• val_ratio: The ratio of validation samples.

• test_ratio: The ratio of test samples.

• num_samples: The number of samples used in random sampling.

• seed: The random seed to use for random sampling. (optional)

Otherwise, in the predefined split method, the task_dict should contain the following keys:

• train_set: The path to the file containing the training set.

• val_set: The path to the file containing the validation set.

• test_set: The path to the file containing the test set.

Other optional keys include:

• num_splits: The number of splits to use for the task. (optional, 1 by default)

Warning

Instantiating a gli.task.GLITask object directly is still experimental. Instead, you should use the gli.task.read_gli_task() or gli.dataloading.get_gli_task() to load tasks from files.

__init__(task_dict, pwd, device='cpu')

Initialize GLITask.

This method will infer whether to use random split or predefined split and then load the split information into GLITask.splits.