`kinoml.datasets.groups`¶

Splitting strategies for datasets

Module Contents¶

class kinoml.datasets.groups.BaseGrouper¶

Base class to assign groups to measurements in a DatasetProvider

assign(dataset, overwrite=False, **kwargs)¶

Given a DatasetProvider, assign a key to the elements of each group, as provided by .indices()

Parameters

dataset (DatasetProvider) –
overwrite (bool, optional=False) – If a measurement has been assigned a group already, do not overwrite unless this option is set to True.

Returns

dataset – The same dataset passed in the input, with measurements modified in place.

Return type

DatasetProvider

abstract indices(dataset, **kwargs)¶

Given a dataset, create a dictionary that maps keys or labels to a set of numerical indices. The strategy to follow will depend on the subclass.

class kinoml.datasets.groups.RandomGrouper(ratios)¶

Randomized groups following a split proportional to the provided ratios

Parameters: ratios (tuple or dict) – 1-based ratios for the different groups. They must sum 1.0. If a dict is provided, the keys are used to label the resulting groups. Otherwise, the groups are 0-enumerated.

indices(dataset, **kwargs)¶

Given a dataset, create a dictionary that maps keys or labels to a set of numerical indices. The strategy to follow will depend on the subclass.

class kinoml.datasets.groups.CallableGrouper(function)¶

A grouper that applies a user-provided function to each Measurement in the Dataset. Returned value should be the name of the group.

Parameters: function (callable) – This function must be able to take a Measurement object and return a str or int.

indices(dataset, progress=True)¶

Given a dataset, create a dictionary that maps keys or labels to a set of numerical indices. The strategy to follow will depend on the subclass.

class kinoml.datasets.groups.BaseFilter¶

Bases: BaseGrouper

Base class to assign groups to measurements in a DatasetProvider