gmmGaussianNumEstimate

Estimate the best number of Gaussians via cross validation

Contents

Syntax

Description

[bestGaussianNum, trainLl, testLl]=gmmGaussianNumEstimate(trainData, testData, gmmOpt) returns the estimated number of Gaussians, together with training and test log likelihood for both the training and test data.

The best number of gaussians is determined according to where the minimum of test log likelihood occurs. (Usually the estimated number of Gaussians is usuallylarger than the desired number of Gaussians, as shown in the example.)

Example

DS=dcData(9);
data=DS.input;
subplot(2,1,1); hist(data, 50);
title('Histogram of the dataset');
trainData=data(:, 1:2:end);
testData=data(:, 2:2:end);
gmmOpt=gmmTrain('defaultOpt');
gmmOpt.arch.gaussianNum=1:2:25;
plotOpt=1;
subplot(2,1,2);
[bestGaussianNum, trainLl, testLl]=gmmGaussianNumEstimate(trainData, testData, gmmOpt, plotOpt);

1/13: No. of Gaussians=1, training LL=-860.012614, test LL=-861.519924
2/13: No. of Gaussians=3, training LL=-788.144759, test LL=-802.060750
3/13: No. of Gaussians=5, training LL=-749.973898, test LL=-767.687532
4/13: No. of Gaussians=7, training LL=-749.706620, test LL=-767.932635
5/13: No. of Gaussians=9, training LL=-745.659400, test LL=-774.328939
6/13: No. of Gaussians=11, training LL=-745.551292, test LL=-773.703669
7/13: No. of Gaussians=13, training LL=-744.003178, test LL=-781.174938
8/13: No. of Gaussians=15, training LL=-741.259317, test LL=-785.536985
9/13: No. of Gaussians=17, training LL=-741.006037, test LL=-784.586421
10/13: No. of Gaussians=19, training LL=-740.398920, test LL=-785.244052
11/13: No. of Gaussians=21, training LL=-740.251365, test LL=-784.997333
12/13: No. of Gaussians=23, training LL=-740.320322, test LL=-785.480987
13/13: No. of Gaussians=25, training LL=-740.414702, test LL=-785.618763

See Also

gmmEval, gmmTrain, gmmInitPrmSet.

Published with MATLAB® R2018a