Contents

Description of demo_binaryclass_exponential.m

Demonstrates two weak base classifiers that can be used in more complex iterative methods like bagging and boosting

clear all
close all
load data_exponential.mat

Broken Stump model

% Train broken stump model
options = [];
[model_stump] = ml_binaryclass_brokenStump(Xtrain,ytrain,options);

% Test broken stump model
yhat = model_stump.predict(model_stump,Xtest);

% Measure test error
testError = sum(yhat~=ytest)/length(ytest);
fprintf('Averaged absolute test error with %s is: %.3f\n', ...
        model_stump.name, testError);

% Plot model predictions
plot2DClassifier(Xtrain,ytrain,model_stump);
Averaged absolute test error with Broken Stump Binary Classification is: 0.136

Exponential-Loss model

% Train exponential loss model
options.addBias = 1;
options.lambdaL2 = 1;
[model_exp] = ml_binaryclass_exponential(Xtrain,ytrain,options);

% Test exponential loss model
yhat = model_exp.predict(model_exp,Xtest);

% Measure test error
testError = sum(yhat~=ytest)/length(ytest);
fprintf('Averaged absolute test error with %s is: %.3f\n', ...
        model_exp.name, testError);

% Plot model predictions
plot2DClassifier(Xtrain,ytrain,model_exp);
Averaged absolute test error with Exponential Loss is: 0.128