Is there any implementation of XGBoost algorithm for decision trees in Matlab?

I've found other boosting algos available in fitensemble and fitcensemble options but not XGBoost. Any chance to find it somewhere else? Thanks

Roberto

[Michelle Hirsch]

MathWorks support team posted a response to this question separately:

https://www.mathworks.com/matlabcentral/answers/457195-are-there-any-xgboost-alternatives-for-matlab

[Bernhard Suhm]

As stated in the article Michelle referred you to, XGBoost is not an algorithm, just an efficient implementation of gradient boosting in Python. MATLAB supports gradient boosting, and since R2019b we also support the binning that makes XGBoost very efficient. You activate the binning with the NumBins name-value parameter to the fit*ensemble functions.

Answer 1

[Jeffrey van Prehn]

Please see: https://nl.mathworks.com/matlabcentral/fileexchange/75898-functions-to-run-xgboost-in-matlab (2 functions to train and test xgboost models). The examples are for classification, but xgboost can also be used for regression. The functions are wrappers for the xgboost.dll library.

[Srishti Gaur]

I am unable to use it. Please hel me out with it.

[Walter Roberson]

@Srishti Gaur what problem did you encounter?

[Srishti Gaur]

Hi Roberson

Here is the error:

Error using movefile

No matching files named 'C:\Post_doc_research\XG_boost\lib\tmp\xgboost\lib\xgboost.dll' were found.

Error in xgboost_install (line 32)

movefile(from, to);

How can I get xgboost.dll file?

Please help me out with this.

[Walter Roberson]

python -m pip install xgboost==1.3.3

should install the dll

Answer 2

[Redha Almahdi]

Hi Roberto,

I am looking for XGBoost matlab based implementation as well. PLease if you get any let me know.

Thanks

[Roberto]

Hi Redha, unfortunately I didn't find any matlab implementation of xgboost so far. But I'm replacing it with the "Method" "AdaBoostM1" that you can find here: https://www.mathworks.com/help/stats/fitcensemble.html#namevaluepairarguments In my tests matlab adaptive boost is outperforming R xgboost in almost any scenario so I'm pretty satisfied. Give it a try!

[Walter Roberson]

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5563301/ talks about preprocessing in MATLAB and about using Python scikit libraries for xgboost. It does not actually state that they call Python from MATLAB but that approach would sound plausible.

[Redha Almahdi]

Thank you Roberto!

Answer 3

[Ali Ebrahimzade]

%% Load Dataset

data = readtable('dataset.csv');

X = data(:,1:end-1); % Input features

y = data(:,end); % Target variable (electrical/thermal efficiency)

%% Split Data into Train and Test

cv = cvpartition(numel(y),'HoldOut',0.2); % 20% for testing

X_train = X(cv.training,:);

y_train = y(cv.training,:);

X_test = X(cv.test,:);

y_test = y(cv.test,:);

%% XGBoost Model

model = XGBTreeBagger('Trees', 200, 'MinLeafSize', 3, 'OOBPrediction','On');

model = fitcensemble(model, X_train, y_train);

y_pred_train = oobPredict(model);

y_pred_test = predict(model, X_test);

%% Extra Trees Model

model = TreeBagger('NumTrees',200,'OOBPredictorImportance','On');

model = fitcensemble(model, X_train, y_train);

y_pred_train = oobPredict(model);

y_pred_test = predict(model, X_test);

%% KNN Model

mdl = fitrknn(X_train,y_train,'NumNeighbors',5);

y_pred_train = predict(mdl,X_train);

y_pred_test = predict(mdl,X_test);

%% Performance Evaluation

R2_train = rsquared(y_train,y_pred_train)

R2_test = rsquared(y_test,y_pred_test)