water.util

Class ModelUtils

java.lang.Object

water.util.ModelUtils

public class ModelUtils
extends java.lang.Object

Shared static code to support modeling, prediction, and scoring.

Used by interpreted models as well as by generated model code.

WARNING: The class should have no other H2O dependencies since it is provided for generated code as h2o-model.jar which contains only a few files.

Field Summary

Fields

Modifier and Type	Field and Description
static float[]	DEFAULT_THRESHOLDS List of default thresholds

Constructor Summary

Constructors

Constructor and Description
ModelUtils()

Method Summary

Methods

Modifier and Type	Method and Description
static float[]	correctProbabilities(float[] scored, float[] priorClassDist, float[] modelClassDist) Correct a given list of class probabilities produced as a prediction by a model back to prior class distribution
static int	getPrediction(float[] preds, double[] data) Utility function to get a best prediction from an array of class prediction distribution.
static int	getPrediction(float[] preds, int row)
static int[]	getPredictions(int numK, float[] preds, double[] data) Create labels from per-class probabilities with pseudo-random tie-breaking, if needed.
static int[]	sampleOOBRows(int nrows, float rate, java.util.Random sampler) Sample out-of-bag rows with given rate with help of given sampler.
static int[]	sampleOOBRows(int nrows, float rate, java.util.Random sampler, int[] oob) In-situ version of sampleOOBRows(int, float, Random).

Methods inherited from class java.lang.Object

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

Field Detail

DEFAULT_THRESHOLDS

public static float[] DEFAULT_THRESHOLDS

List of default thresholds

Constructor Detail

ModelUtils

public ModelUtils()

Method Detail

getPrediction

public static int getPrediction(float[] preds,
                double[] data)

Utility function to get a best prediction from an array of class prediction distribution. It returns index of max value if predicted values are unique. In the case of tie, the implementation solve it in pseudo-random way.

Parameters:

preds - an array of prediction distribution. Length of arrays is equal to a number of classes+1.

Returns:

the best prediction (index of class, zero-based)

getPredictions

public static int[] getPredictions(int numK,
                   float[] preds,
                   double[] data)

Create labels from per-class probabilities with pseudo-random tie-breaking, if needed.

Parameters:

numK - Number of top probabilities to make labels for

preds - Predictions (first element is ignored here: placeholder for a label)

data - Data to break ties (typically, the test set data for this row)

Returns:

Array of predicted labels

getPrediction

public static int getPrediction(float[] preds,
                int row)

correctProbabilities

public static float[] correctProbabilities(float[] scored,
                           float[] priorClassDist,
                           float[] modelClassDist)

Correct a given list of class probabilities produced as a prediction by a model back to prior class distribution

The implementation is based on Eq. (27) in the paper.

Parameters:

scored - list of class probabilities beginning at index 1

priorClassDist - original class distribution

modelClassDist - class distribution used for model building (e.g., data was oversampled)

Returns:

corrected list of probabilities

sampleOOBRows

public static int[] sampleOOBRows(int nrows,
                  float rate,
                  java.util.Random sampler)

Sample out-of-bag rows with given rate with help of given sampler. It returns array of sampled rows. The first element of array contains a number of sampled rows. The returned array can be larger than number of returned sampled elements.

Parameters:

nrows - number of rows to sample from.

rate - sampling rate

sampler - random "dice"

Returns:

an array contains numbers of sampled rows. The first element holds a number of sampled rows. The array length can be greater than number of sampled rows.

sampleOOBRows

public static int[] sampleOOBRows(int nrows,
                  float rate,
                  java.util.Random sampler,
                  int[] oob)

In-situ version of sampleOOBRows(int, float, Random).

Parameters:

oob - an initial array to hold sampled rows. Can be internally reallocated.

Returns:

an array containing sampled rows.

See Also:

sampleOOBRows(int, float, Random)