smile.validation

Class GroupKFold

java.lang.Object

smile.validation.GroupKFold

public class GroupKFold
extends java.lang.Object

GroupKfold is a cross validation technique that splits the data by respecting additional information about groups. Each sample belongs to a certain group and each group can have multiple samples. The goal of the split is then to assure that no samples from the same group are present in the training and testing subsets in a single fold. This is useful when the i.i.d. assumption is known to be broken by the underlying process generating the data. E.g. when we have multiple samples by the same user and want to make sure that the model learns the concept we are interested in rather than user-specific features that don't generalize to unseen users, this approach could be used. The implementation is inspired by scikit-learn's implementation of sklearn.model_selection.GroupKFold.

Field Summary

Fields

Modifier and Type	Field and Description
int	k The number of folds.
int[][]	test The index of testing instances.
int[][]	train The index of training instances.

Constructor Summary

Constructors

Constructor and Description
GroupKFold(int n, int k, int[] groups) Constructor.

Method Summary

Modifier and Type	Method and Description
int[]	classification(smile.data.DataFrame data, java.util.function.Function<smile.data.DataFrame,DataFrameClassifier> trainer) Runs cross validation tests.
<T> int[]	classification(T[] x, int[] y, java.util.function.BiFunction<T[],int[],Classifier<T>> trainer) Runs cross validation tests.
double[]	regression(smile.data.DataFrame data, java.util.function.Function<smile.data.DataFrame,DataFrameRegression> trainer) Runs cross validation tests.
<T> double[]	regression(T[] x, double[] y, java.util.function.BiFunction<T[],double[],Regression<T>> trainer) Runs cross validation tests.

Methods inherited from class java.lang.Object

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

Field Detail

k

public final int k

The number of folds.

train

public final int[][] train

The index of training instances.

test

public final int[][] test

The index of testing instances.

Constructor Detail

GroupKFold

public GroupKFold(int n,
                  int k,
                  int[] groups)

Constructor.

Parameters:

n - the number of samples.

k - the number of folds.

groups - the group information of samples.

Method Detail

classification

public <T> int[] classification(T[] x,
                                int[] y,
                                java.util.function.BiFunction<T[],int[],Classifier<T>> trainer)

Runs cross validation tests.

Returns:

the predictions.

classification

public int[] classification(smile.data.DataFrame data,
                            java.util.function.Function<smile.data.DataFrame,DataFrameClassifier> trainer)

Runs cross validation tests.

Returns:

the predictions.

regression

public <T> double[] regression(T[] x,
                               double[] y,
                               java.util.function.BiFunction<T[],double[],Regression<T>> trainer)

Runs cross validation tests.

Returns:

the predictions.

regression

public double[] regression(smile.data.DataFrame data,
                           java.util.function.Function<smile.data.DataFrame,DataFrameRegression> trainer)

Runs cross validation tests.

Returns:

the predictions.