This example shows how to create and compare kernel approximation classifiers in the Classification Learner app, and export trained models to the workspace to make predictions for new data. You can use kernel approximation classifiers to perform nonlinear classification of data with many observations. For large in-memory data, kernel classifiers tend to train and predict faster than SVM classifiers with Gaussian kernels.
In the MATLAB® Command Window, load the
set, and create a table from the variables in the data set to use for
classification. The data set contains 24,075 observations of five physical human
activities: sitting, standing, walking, running, and dancing. Each observation
has 60 features extracted from acceleration data measured by smartphone
load humanactivity Tbl = array2table(feat); Tbl.Properties.VariableNames = featlabels'; activity = categorical(actid,1:5,actnames); Tbl.Activity = activity;
Alternatively, you can load the
humanactivity data set,
create the categorical
activity response variable, and keep
activity data as separate
Click the Apps tab, and then click the Show more arrow on the right to open the apps gallery. In the Machine Learning and Deep Learning group, click Classification Learner.
On the Classification Learner tab, in the File section, click New Session and select From Workspace.
In the New Session from Workspace dialog box, select the table
Tbl from the Data Set Variable list.
Note that the app selects response and predictor variables based on their data
types. In particular, the app selects
Activity as the
response variable because it is the only categorical variable. For this example,
do not change the selections.
Alternatively, if you keep the predictor data
activity as two separate variables, you can
first select the matrix
feat from the Data Set
Variable list. Then, under Response, click
the From workspace option button and select
activity from the list.
To accept the default validation scheme and continue, click Start Session. The default validation option is 5-fold cross-validation, to protect against overfitting.
Classification Learner creates a scatter plot of the data.
Use the scatter plot to investigate which variables are useful for predicting the response. Select different options in the X and Y lists under Predictors to visualize the distribution of activities and measurements. Note which variables separate the activities (colors) most clearly.
Create a selection of kernel approximation models. On the Classification Learner tab, in the Model Type section, click the arrow to open the gallery. In the Kernel Approximation Classifiers group, click All Kernels.
In the Training section, click Train.
If you have Parallel Computing Toolbox™, you can train all the models (All Kernels) simultaneously by selecting the Use Parallel button in the Training section before clicking Train. After you click Train, the Opening Parallel Pool dialog box opens and remains open while the app opens a parallel pool of workers. During this time, you cannot interact with the software. After the pool opens, the app trains the models simultaneously.
Classification Learner trains one of each kernel approximation option in the gallery. In the Models pane, the app outlines the Accuracy (Validation) score of the best model. Classification Learner also displays a validation confusion matrix for the first kernel model (SVM Kernel).
Validation introduces some randomness into the results. Your model validation results can vary from the results shown in this example.
To view the results for a model, select the model in the Models pane, and inspect the Current Model Summary pane. The Current Model Summary pane displays the Training Results metrics, calculated on the validation set.
Select the second kernel model (Logistic Regression Kernel) in the Models pane, and inspect the accuracy of the predictions in each class using a validation confusion matrix. On the Classification Learner tab, in the Plots section, click the arrow to open the gallery, and then click Confusion Matrix (Validation) in the Validation Results group. View the matrix of true class and predicted class results.
Compare the confusion matrices for the two models side-by-side. On the Classification Learner tab, in the Plots section, click the Layout button and select Compare models. In the top right of each plot, click the Hide plot options button to make more room for the plot.
To return to the original layout, you can click the Layout button in the Plots section and select Single model (Default).
Choose the best model in the Models pane (the best score is highlighted in the Accuracy (Validation) box). To improve the model, try including different features in the model. See if you can improve the model by removing features with low predictive power.
To investigate features to include or exclude, use the scatter and parallel coordinates plots. On the Classification Learner tab, in the Plots section, open the gallery and click the Scatter and Parallel Coordinates buttons in the Validation Results group.
To specify which features to include, click Feature Selection in the Features section on the Classification Learner tab. In the Feature Selection dialog box, specify predictors to remove from the model and click OK. In the Training section, click Train to train a new model using the new options. Compare results among the classifiers in the Models pane.
Choose the best model in the Models pane. To try to improve the model further, change its advanced settings. On the Classification Learner tab, in the Model Type section, click Advanced and select Advanced. In the Advanced Kernel Options dialog box, try changing some of the settings and click OK. Train the new model by clicking Train in the Training section.
To learn more about kernel model settings, see Kernel Approximation Classifiers.
You can export a compact version of the trained model to the workspace. On the Classification Learner tab, in the Export section, click Export Model and select either Export Model or Export Compact Model. Note that either option exports a compact version of the trained model because kernel approximation models do not store training data. See Export Classification Model to Predict New Data.
To examine the code for training this classifier, click Generate
Function in the Export section. Because the
data set used to train this classifier has more than two classes, the generated
code uses the
fitcecoc function rather than
Use the same workflow to evaluate and compare the other classifier types you can train in Classification Learner.
To train all the nonoptimizable classifier model presets available for your data set:
On the Classification Learner tab, in the Model Type section, click the arrow to open the gallery of models.
In the Get Started group, click All. Then, in the Training section, click Train.
To learn about other classifier types, see Train Classification Models in Classification Learner App.