org.openimaj.experiment.validation.cross

Class GroupedKFold<KEY,INSTANCE>

java.lang.Object

org.openimaj.experiment.validation.cross.GroupedKFold<KEY,INSTANCE>

Type Parameters:

KEY - Type of groups

INSTANCE - Type of instances

All Implemented Interfaces:

CrossValidator<GroupedDataset<KEY,ListDataset<INSTANCE>,INSTANCE>>

public class GroupedKFold<KEY,INSTANCE>
extends Object
implements CrossValidator<GroupedDataset<KEY,ListDataset<INSTANCE>,INSTANCE>>

K-Fold Cross-Validation on grouped datasets.

All the instances are split into k subsets. The validation data in each iteration is one of the subsets, whilst the training data is the remaindering subsets. The subsets are not guaranteed to have any particular balance of groups as the splitting is completely random; however if there is the same number of instances per group, then the subsets should be balanced on average. A particular fold could potentially have no training or validation data for a particular class.

Setting the number of splits to be equal to the number of total instances is equivalent to LOOCV. If LOOCV is the aim, the GroupedLeaveOneOut class is a more efficient implementation than this class.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

Constructor Summary

Constructors

Constructor and Description
GroupedKFold(int k) Construct the GroupedKFold with the given number of folds.

Method Summary

Modifier and Type	Method and Description
CrossValidationIterable<GroupedDataset<KEY,ListDataset<INSTANCE>,INSTANCE>>	createIterable(GroupedDataset<KEY,ListDataset<INSTANCE>,INSTANCE> data) Create the iterable from the dataset.
String	toString()

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

GroupedKFold

public GroupedKFold(int k)

Construct the GroupedKFold with the given number of folds.

Parameters:

k - the target number of folds.

Method Detail

createIterable

public CrossValidationIterable<GroupedDataset<KEY,ListDataset<INSTANCE>,INSTANCE>> createIterable(GroupedDataset<KEY,ListDataset<INSTANCE>,INSTANCE> data)

Description copied from interface: CrossValidator

Create the iterable from the dataset.

Specified by:

createIterable in interface CrossValidator<GroupedDataset<KEY,ListDataset<INSTANCE>,INSTANCE>>

Parameters:

data - the dataset

Returns:

the iterable

toString

public String toString()

Overrides:

toString in class Object