Code Generation for Binary GLM Logistic Regression Model Trained in Classification Learner

This example shows how to train a binary GLM logistic regression model using Classification Learner, and then generate C code that predicts labels using the exported classification model.

Load Sample Data

Load sample data and import the data into the Classification Learner app.

Load the patients data set. Specify the predictor data X, consisting of p predictors, and the response variable Y.

load patients
X = [Age Diastolic Height Systolic Weight];
p = size(X,2);
Y = Gender;

On the Apps tab, click the Show more arrow at the right of the Apps section to display the gallery, and select Classification Learner. On the Classification Learner tab, in the File section, select New Session > From Workspace.

In the New Session from Workspace dialog box, under Data Set Variable, select X from the list of workspace variables. Under Response, click the From workspace option button and then select Y 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.

By default, Classification Learner creates a scatter plot of the data.

Train Binary GLM Logistic Regression Model

Train a binary GLM logistic regression model using the Classification Learner app.

On the Classification Learner tab, in the Models section, click the Show more arrow to display the gallery of classifiers. Under Logistic Regression Classifiers, click Binary GLM Logistic Regression. In the Train section, click Train All and select Train Selected. The app trains the model and displays its cross-validation accuracy Accuracy (Validation) in the Models pane.

Export Model to Workspace

Export the model to the MATLAB® Workspace and save it using saveLearnerForCoder.

On the Classification Learner tab, click Export, click Export Model, and select Export Model. In the dialog box, specify trainedLogisticRegressionModel as the model name and click OK.

The structure trainedLogisticRegressionModel appears in the MATLAB Workspace. The field GeneralizedLinearModel of trainedLogisticRegressionModel contains the required model.

Note: If you run this example with all supporting files, you can load the trainedLogisticRegressionModel.mat file at the command line rather than exporting the model. The trainedLogisticRegressionModel structure was created using the previous steps.

load('trainedLogisticRegressionModel.mat')

At the command line, save the model to a file named myModel.mat in your current folder.

saveLearnerForCoder(trainedLogisticRegressionModel.GeneralizedLinearModel,'myModel')

Additionally, save the names of the success, failure, and missing classes of the trained model.

classNames = {trainedLogisticRegressionModel.SuccessClass, ...
    trainedLogisticRegressionModel.FailureClass,trainedLogisticRegressionModel.MissingClass};
save('ModelParameters.mat','classNames');

Generate C Code for Prediction

Define the entry-point function for prediction, and generate code for the function by using codegen.

In your current folder, define a function named classifyX.m that does the following:

• Accepts a numeric matrix (X) of observations containing the same predictor variables as the ones used to train the logistic regression model

• Loads the classification model in myModel.mat

• Computes predicted probabilities using the model

• Converts the predicted probabilities to indices, where 1 indicates a success, 2 indicates a failure, and 3 indicates a missing value

• Loads the class names in ModelParameters.mat

• Returns predicted labels by indexing into the class names

function label = classifyX (X) %#codegen 
%CLASSIFYX Classify using Logistic Regression Model 
%  CLASSIFYX classifies the measurements in X 
%  using the logistic regression model in the file myModel.mat, 
%  and then returns class labels in label.

n = size(X,1);
label = coder.nullcopy(cell(n,1));

Mdl = loadLearnerForCoder('myModel');
probability = predict(Mdl,X);

index = ~isnan(probability).*((probability<0.5)+1) + isnan(probability)*3;

classInfo = coder.load('ModelParameters');
classNames = classInfo.classNames;

for i = 1:n    
    label{i} = classNames{index(i)};
end
end

Note: If you create a logistic regression model in Classification Learner after using feature selection or principal component analysis (PCA), you must include additional lines of code in your entry-point function. For an example that shows these additional steps, see Code Generation and Classification Learner App.

Generate a MEX function from classifyX.m. Create a matrix data for code generation using coder.typeof. Specify that the number of rows in data is arbitrary, but that data must have p columns, where p is the number of predictors used to train the logistic regression model. Use the -args option to specify data as an argument.

data = coder.typeof(X,[Inf p],[1 0]);
codegen classifyX.m -args data

Code generation successful.

codegen generates the MEX file classifyX_mex.mex64 in your current folder. The file extension depends on your platform.

Verify that the MEX function returns the expected labels. Randomly draw 15 observations from X.

rng('default') % For reproducibility
testX = datasample(X,15);

Classify the observations by using the predictFcn function in the classification model trained in Classification Learner.

testLabels = trainedLogisticRegressionModel.predictFcn(testX);

Classify the observations by using the generated MEX function classifyX_mex.

testLabelsMEX = classifyX_mex(testX);

Compare the sets of predictions. isequal returns logical 1 (true) if testLabels and testLabelsMEX are equal.

isequal(testLabels,testLabelsMEX)

ans = logical
   1

predictFcn and the MEX function classifyX_mex return the same values.