View the tree ensemble trained by fitcensemble

Dear, let's suppose I'm training a bagged tree ensemble on the 40 sampe points (29 features per sample)

The GT labels are in the var labels.

I gat a tree mdlRF as follows:

dataSel = rand(40, 29);  % just a toy example
labels = logical(randi(1,29,1)); % just a toy example

t = templateTree('PredictorSelection','interaction-curvature','Surrogate','on', ...
    'Reproducible',true); % For reproducibility of random predictor selections
 mdlRF = fitcensemble(dataSel,labels,'Method','Bag', ...
    'CategoricalPredictors',[false true true true true true], 'PredictorNames', vars,...
    'NumLearningCycles',30,'Learners',t);

If the trained learner were a simple tree I could have viewed it in graph /text mode with the command:

view(mdlRF, 'Mode', 'graph')

Is there an alternative way to view how the mdlRF makes the decision??

Answer 1

[Puru Kathuria]

Hi,

I understand that you want to visualize your ensemble after training it on a dataset. You can try replacing 2nd line of code with the following line.

labels = logical(randi(1,40,1)); % Training data points (X) should be equal to training labels (Y)

And the last line with the following line.

view(mdlRF.Trained{1}.CompactRegressionLearner,'Mode','graph'); % Visualising 

This might meet your requirements and help you visualize your model.

[Elena Casiraghi]

Sorry Puru!

The code was wrong in the number of labels

dataSel = rand(40, 29);  % just a toy example: 40 points
labels = logical(randi(1,40,1)); % just a toy example: 40 labels
t = templateTree('PredictorSelection','interaction-curvature','Surrogate',1, ...
    'Reproducible',true); 
% For reproducibility of random predictor selections
% Use surrogate = 1 to generate surrogate 1 branch for each split: when one value is NaN in the data, the surrogate branch i used 


numTrees = 7;
% train a random forest with 7 trees
mdlRF = fitcensemble(dataSel,labels,'Method','Bag', ...
    'CategoricalPredictors',[false true true true true true], 'PredictorNames', vars,...
    'NumLearningCycles',numTrees,'Learners',t);


for numTree = 1: numTrees
    view(mdlRF.Trained{1}, 'Mode', 'graph');
end

This show all the seven trees, BUT it does not show the alternative (surrogate) branches that are used when a Nan Value is present.

How can I do that?