How to set tree arguments in TreeBagger

Hi.

I'm using TreeBagger for classification and I want to control the depth of the trees in it.

I have read that you can control the depth of the trees using tree parameters, MinLeafSize, MinParentSize and MinParentSize in here https://la.mathworks.com/help/stats/fitctree.html#bt6cr7t_sep_shared-MaxNumSplits.

Also in the TreeBagger documentation says that it has an argument TreeArguments that receive a Cell array of arguments for fitctree or fitrtree. These arguments are used by TreeBagger when growing new trees for the ensemble.

I'm trying to set this property but I get the next error with the next code.

    maxNumSplits = 3; %Maximal number of decision splits, positive integer
    minLeafSize = 3; %Mini um number of leaf node observations, positive integer
    minParentSize = 3; %Minimum number of brach node observation, positive integer
    numberPredictorsToSample = 3; %number of random feature for each decision, default = squareroot of the total number of features
    numTrees = 500; %Scalar value equal to the number of desicion trees

    treeArguments = {'MaxNumSplits',maxNumSplits,'MinLeafSize',minLeafSize,'MinParentSize',minParentSize};

    RF = TreeBagger(numTrees,trainingData,trainingClass,'TreeArguments',treeArguments);

The Error I get is

Error using internal.stats.parseArgs (line 42) Invalid parameter name: TreeArguments.

Error in classreg.learning.FullClassificationRegressionModel.prepareDataCR (line 115) [W,predictornames,responsename,catpreds] = ...

Error in ClassificationTree.prepareData (line 471) [X,Y,W,dataSummary] = ...

Error in TreeBagger/init (line 1190) [bagger.X,y,bagger.W,bagger.DataSummary,classSummary] = ...

Error in TreeBagger (line 531) bagger = init(bagger,X,Y,makeArgs{:});

Error in variabilidad (line 42) RF = TreeBagger(numTrees,trainingData,trainingClass,'TreeArguments',templateTree);

Answer 1

[luis barbosa]

Dear Fabian,

I saw your code and there are some errors there.

First, the TreeBagger (function/class) does not accept as argument 'TreeArguments'. So it is not possible to creat a cell-array (the case of your code), or a structure and simple pass it to the Trebagger as an argument.

One way to run your code is the following:

    maxNumSplits = 3; %Maximal number of decision splits, positive integer
    minLeafSize = 3; %Mini um number of leaf node observations, positive integer
    minParentSize = 3; %Minimum number of brach node observation, positive integer
    numberPredictorsToSample = 3; %number of random feature for each decision, default = squareroot of the total number of features
    numTrees = 500; %Scalar value equal to the number of desicion trees
    method = 'regression'; %or 'classification' 
    
    %% training a TreeBagger.
    % It is a random forests since the numberPredictorsToSample is less
    % than the amount of inputs in the trainingData, used to train the RF.
    % Otherwise, we have a simple bagged ensemble of trees.
    
    % assuming:
    %    -  trainingData is a table
    %    -  y_label is a string with the output name
    
    RF = TreeBagger(numTrees,trainingData,y_label,...    
        'Method',method,...        %you have to say whether it is a regression or classification problem
        'MaxNumSplits',maxNumSplits,...
        'MinLeafSize',minLeafSize,...
        'NumPredictorsToSample',numberPredictorsToSample);

I left out the 'MinParentSize', because an error occurred, namely:

"Error using TreeBagger/init (line 1414)

TreeBagger does not accept 'MinParentSize' and 'SplitMin' input parameters. Use 'MinLeafSize'."

I didn't try to solve this problem. Perhaps, a possible solution is to set this parameter by using templateTree, and pass it to TreeBagger, something like this:

    t = templateTree('MinParentSize',number);
    RF = TreeBagger(numTrees,trainingData,y_label,...    
        'Method',method,...        %you have to say whether it is a regression or classification problem
        'MaxNumSplits',maxNumSplits,...
        'MinLeafSize',minLeafSize,...
        'NumPredictorsToSample',numberPredictorsToSample,...
        'Learners',t); %passing the created templateTree object