Core-set selection

TlseHypDataSet.utils.coreset.core_set_selection(dataset, budget=10, metric=None, dim_reduction='autoencoder', n_components=8)

Performs a class-wise core-set selection with a k-center greedy algorithm as implemented in https://github.com/PatrickZH/DeepCore

Parameters

• dataset – a torch.utils.data.Dataset object

• budget – number of samples to select by class

• metric – distance metric to use (a callable)

• dim_reduction – method to represent the data in a lower dimensional space (‘pca’ or ‘autoencoder’)

• n_components – dimension of the representation space

Returns

A core-set selection of the data

TlseHypDataSet.utils.coreset.random_selection(dataset, budget=10)

Performs a class-wise random selection

Parameters

• dataset – a torch.utils.data.Dataset object

• budget – number of samples to select by class

Returns

A random selection of the data