how to change crossvalind to cvpartition

Hello,

What changes should i make to below code, to use cvpartition for cross validation?

TREES = [2 4 6 8 10 20 40:20:80 100:50:300 400 500];

FEATURES = [1:size(X1,2)]; % Breiman's rule: round(sqrt(size(X, 2)),0)

grid_AUC_crossval = zeros(length(TREES), length(FEATURES)); % to store train AUC scores

grid_F1_crossval = zeros(length(TREES), length(FEATURES));

for t=1:length(TREES)

for f=1:length(FEATURES)

trees = TREES(t);

features = FEATURES(f);

% run cross-validation on every model iteration

numFolds = 10;

Indices = crossvalind('Kfold', y1, numFolds);

final_preds = [];

final_scores = [];

yT = [];

for i = 1:numFolds

X2_fold = X1(Indices == i, :);

X1_fold = X1(Indices ~= i, :);

y1_fold = y1(Indices ~= i, :);

testIdx = (Indices == i); % index numbers of test items

Mdl = TreeBagger(trees, X1_fold, y1_fold, 'NumPredictorsToSample', features,'MinLeafSize', 5, 'Method', 'classification');

[preds, scores] = predict(Mdl, X2_fold);

Answer 1

[Anmol Dhiman]

cvpartition is used for creating cross validation partition of the data.

You can use in the method as shown below

TREES = [2 4 6 8 10 20 40:20:80 100:50:300 400 500]; 
FEATURES = [1:size(X1,2)];  % Breiman's rule: round(sqrt(size(X, 2)),0)
X1 = meas;
grid_AUC_crossval = zeros(length(TREES), length(FEATURES)); % to store train AUC scores
grid_F1_crossval =  zeros(length(TREES), length(FEATURES));
 y1 = species;
for t=1:length(TREES)
    for f=1:length(FEATURES)
        trees = TREES(t);
        features = FEATURES(f);
        
        % run cross-validation on every model iteration
        numFolds = 10;

        c = cvpartition(y1,'KFold',numFolds);
        final_preds = [];
        final_scores = [];
        yT = [];
        
        for i = 1:numFolds
            idx = training(c,i); % get indices for all trainings data
            testIdx = (idx ~= i); % index numbers of test items
            X2_fold = X1(idx ~= i, :);
            X1_fold = X1(idx == i, :);
            y1_fold = y1(idx == i, :);
                  
             Mdl = TreeBagger(trees, X1_fold, y1_fold, 'NumPredictorsToSample', features,'MinLeafSize', 5,  'Method', 'classification');
            [preds, scores] = predict(Mdl, X2_fold);
        end
    end
end

[kav]

Thank you for your answer.

but i am getting below error,not able to understand where i am going wrong.

Error in TreeBagger/init (line 1338)

ClassificationTree.prepareData(x,y,...

Error in TreeBagger (line 616)

bagger = init(bagger,X,Y,makeArgs{:});

Error in cwcrossvalind (line 64)

Mdl = TreeBagger(trees, X1_fold, y1_fold, 'NumPredictorsToSample', features,'MinLeafSize', 5,

'Method', 'classification');

[Anmol Dhiman]

Try using this

 for i = 1:numFolds
            idx = training(c,i); % get indices for all trainings data
            testIdx = (idx ~= i); % index numbers of test items
            X2_fold = X1(idx ~= 1, :);
            X1_fold = X1(idx == 1, :);
            y1_fold = y1(idx == 1, :);

[kav]

Thank you Anmol.

Program is running from long time, to make it quicker what changes should i make it?

is it with number of trees?

[Anmol Dhiman]

You may change the number of tree to check the accuracies that you are getting, besides it will take less time for getting the results if you use less number of trees. Also you may try using parfor instead of for loop for decreasing the total time.