ImportanceIndexWrapper

class axtreme.data.importance_dataset.ImportanceIndexWrapper(dataset: Dataset[T_np_interface], importance_idx: int)

Bases: Dataset[tuple[T_np_interface, float]]

Wraps an existing dataset, returning a column/index of the item as the importance weight.

__init__(dataset: Dataset[T_np_interface], importance_idx: int) → None

Wrap an existing dataset, returning part of the item as the importance weight.

Parameters:

• dataset – A dataset, where one of the columns should be used as an importance weight. - Note: Currently only datasets that return numpy or torch tensors are supported

• importance_idx – the column index containing the importance weights.

Todo

• Generalise this to deal with other types of dataset output (list, numpy etc).

Methods

__init__(dataset, importance_idx)

Wrap an existing dataset, returning part of the item as the importance weight.