Export Partitions and Data Sets from Classification Learner or Regression Learner

Since R2026a

This example shows how to export the partitions used to compute validation and test metrics, as well as the training and test data sets, from the current Classification Learner or Regression Learner session to a structure variable in the MATLAB® workspace. You can use the exported structure to reproduce the training results outside Classification Learner and Regression Learner. For more information about validation partitions and test data sets, see Select Validation Scheme in Classification Learner or Regression Learner and Test Trained Models in Classification Learner or Regression Learner.

Export Partitions and Data Sets from App Session

You can export partitions and data sets from your current session in either the Classification Learner App or the Regression Learner App.

  1. In the MATLAB Command Window, load the carbig data set, which contains measurements of cars made in the 1970s and early 1980s.

    load carbig

  2. Create a table containing the predictor variables Acceleration, Displacement, Horsepower, and MPG, as well as the response variable Cylinders.

    cars = table(Acceleration,Displacement,Horsepower,MPG,Cylinders);

  3. On the Apps tab, in the Machine Learning and Deep Learning group, click Classification Learner or Regression Learner.

  4. On the Learn tab, in the File section, select New Session > From Workspace Data.

  5. In the New Session from Workspace Data dialog box, select the cars table from the Data Set Variable list. The app selects the response and predictor variables. The default response variable is Cylinders. The default validation scheme is 5-fold cross-validation, to protect against overfitting.

    In the Test section of the dialog box, select the check box to set aside a test data set. The default option sets aside 10 percent of the imported data.

  6. To accept the options and continue, click Start Session.

  7. In the Export section of the Learn tab, select Export > Export Partitions and Data Sets. The Export Partitions and Data Sets dialog box opens.

    Export Partitions and Data Sets dialog box

  8. In the Variable name box, edit the name of the exported variable, if necessary. By default, the exported structure includes the validation partition, training (and validation) data, and test data set from the current session.

  9. Because you chose to set aside a portion of the original data set as a test data set at the start of the session, you can additionally include the test partition and the original data set in the exported structure. Click Additional options and select both check boxes.

  10. Click OK to export the structure to the workspace.

    The exported structure contains the following fields:

    • ValidationPartition — cvpartition object that defines the partitions of the current session's training data set that are used to compute validation metrics

    • TrainingDataSet — Structure that contains the current session's training data set, which is used to train the final version of each model.

    • TestDataSet — Structure that contains the current session's test data set

    • TestPartition — cvpartition object that defines the partition of the original data set (from the start of the session) which you set aside to create the test data set

    • OriginalDataSet — Structure that contains the original data set from the start of the session, prior to setting aside the test data set

    • About — String that contains information about the app and MATLAB version used to export the structure

Extract Data Sets from Exported Structure Variable

You can use the exported structure variable partitionsAndDataSets to extract the training, validation, and test data sets from the app session at the MATLAB command line.

Extract Training Data Set

Display the training data set structure.

partitionsAndDataSets.TrainingDataSet
ans = struct with fields:
              Data: [366×5 table]
    PredictorNames: {'Acceleration'  'Displacement'  'Horsepower'  'MPG'}
      ResponseName: {'Cylinders'}
Because the original data set has 406 observations, and you set aside 10% as a test data set at the start of the session, the training data set contains 366 observations.

Create a table variable named trainingSet that contains the training data.

trainingSet = partitionsAndDataSets.TrainingDataSet.Data;

Extract Validation Fold Data Sets

Display the validation partition object properties and values.

partitionsAndDataSets.ValidationPartition

ans = 
K-fold cross validation partition
    NumObservations: 366
        NumTestSets: 5
          TrainSize: [293 292 293 293 293]
           TestSize: [73 74 73 73 73]
           IsCustom: 0
          IsGrouped: 0
       IsStratified: 0


  Properties, Methods
The NumObservations value is equal to the number of training data set observations. The NumTestSets value indicates that the partition object has five cross-validation folds. The TestSize value contains the number of observations from the training data set that are in each fold. For more information about cross-validation folds, see Select Validation Scheme in Classification Learner or Regression Learner.

Use the test function to create an array named index1 containing the row indices of the training data set that are in the first validation fold.

index1 = test(partitionsAndDataSets.ValidationPartition,1);
Create a table named validationFold1 that contains the data in the first validation fold.
validationFold1 = trainingSet(index1,:);

Extract Test Data Set

Display the test data set structure.

partitionsAndDataSets.TestDataSet

ans = struct with fields:
              Data: [40×5 table]
    PredictorNames: {'Acceleration'  'Displacement'  'Horsepower'  'MPG'}
      ResponseName: {'Cylinders'}
Because the original data set has 406 observations, and you set aside 10% as a test data set at the start of the session, the training data set contains 366 observations.

Create a table variable named testSet that contains the test data set.

testSet = partitionsAndDataSets.TestDataSet.Data;

Extract Original Data Set and Test Data from Start of Session

The TestDataSet field of partitionsAndDataSets contains the current test data set when you export the structure variable. If you set aside a test data set when you start the session, and then load a new test data set during the session, you can extract the original data set and test data set from the start of the session.

Create a table originalData that contains the original data set from the start of the session (prior to setting aside the test data set).

originalData = partitionsAndDataSets.OriginalDataSet.Data;
Create an array named indexTest containing the row indices of the original data set that were set aside as test data at the start of the session.
indexTest = test(partitionsAndDataSets.TestPartition);
Create a table named originalTestSet containing the data set aside for testing at the start of the session.
originalTestSet = originalData(indexTest,:);
In this example, the variables testSet and originalTestSet are identical, because you did not load a new test data set during the session.

See Also

Topics