unrecognized table variable name minLS in Bayesian optimization of TreeBagger

Hello,

I tried to do the hyperparameter tuning using bayesian optimization for the randomforest model I made using Treebagger. This is the code I used. I didnot use ensemble bagged trees in regression learner directly because eachtime i check the optimization the minimum mse output is like 498 learners and 1 minimum leaf size for 9 number of predictors.

264    inputTable=readtable('dataall_trainingregression.csv');
265    predictorNames = {'temp_diff', 'temp_median', 'NDVI', 'Clay', 'elevation', 'slope', 'TWI', 'sand', 'DOY'};
266    predictors = inputTable(:, predictorNames);
267    response = inputTable.daily_meanSM;
268    n=length(inputTable.daily_meanSM);
269    cvp = cvpartition(n,'KFold',5);
    
271    maxMinLS = 20;
272    minLS = optimizableVariable('minLS',[1,maxMinLS],'Type','integer');
273    numPTS = optimizableVariable('numPTS',[1,size(predictors,2)],'Type','integer');
274    hyperparametersRF = [minLS; numPTS];
275    fun = @(params)crossval('mse',predictors,response,'Predfun',@myfunction,'Partition',cvp);
    
277    results = bayesopt(fun,hyperparametersRF,...
        'AcquisitionFunctionName','expected-improvement-plus','Verbose',0);
    
280    function yfit = myfunction(params,predictors,response,test)
281    Mdl1 = TreeBagger(30,predictors,response,...
282        'Method',"regression",'Surrogate',"on",...
283        'PredictorSelection',"curvature",...
284        'OOBPredictorImportance',"on",'MinLeafSize',params.minLS,...
285       'NumPredictorsToSample',params.numPTS);
286    yfit = predict(Mdl1,test);
    
288    end

for this code i receive an error message of :

Error using crossval>evalFun

The function 'myfunction' generated the following error:

Unrecognized table variable name 'minLS'.

Error in crossval>getLossVal (line 529)

funResult = evalFun(funorStr,arg(1:end-1));

Error in crossval (line 428)

[funResult,outarg] = getLossVal(i, nData, cvp, data, predfun);

Error in model_hyperparameter_tuning>@(params)crossval('mse',predictors,response,'Predfun',@myfunction,'Partition',cvp) (line 275)

fun = @(params)crossval('mse',predictors,response,'Predfun',@myfunction,'Partition',cvp);

Error in BayesianOptimization/callObjNormally (line 13)

Objective = this.ObjectiveFcn(conditionalizeX(this, X));

Error in BayesianOptimization/callObjFcn (line 25)

= callObjNormally(this, X);

Error in BayesianOptimization/runSerial (line 24)

ObjectiveFcnObjectiveEvaluationTime, ObjectiveNargout] = callObjFcn(this, this.XNext);

Error in BayesianOptimization/run (line 9)

this = runSerial(this);

Error in BayesianOptimization (line 184)

this = run(this);

Error in bayesopt (line 323)

Results = BayesianOptimization(Options);

Error in model_hyperparameter_tuning (line 277)

results = bayesopt(fun,hyperparametersRF,...

can anyone please help me to solve this problem. Thanks

Answer 1

[Jack]

Hey Kalani,

The error "Unrecognized table variable name 'minLS'" happens because MATLAB is trying to access params.minLS as if params were a table, but in your function myfunction, params is actually a struct, not a table.

Fix: Access params as a struct

Change this line in myfunction:

Mdl1 = TreeBagger(30,predictors,response,...
    'Method',"regression",'Surrogate',"on",...
    'PredictorSelection',"curvature",...
    'OOBPredictorImportance',"on",'MinLeafSize',params.minLS,...
    'NumPredictorsToSample',params.numPTS);

to

Mdl1 = TreeBagger(30, predictors, response,...
    'Method', "regression", 'Surrogate', "on",...
    'PredictorSelection', "curvature",...
    'OOBPredictorImportance', "on", 'MinLeafSize', params.minLS,...
    'NumPredictorsToSample', params.numPTS);

Alternative Fix:

If the error persists, another way to reference params safely is to convert it to a structure explicitly:

params = table2struct(params);

This ensures MATLAB treats params as a struct with fields minLS and numPTS.

This should resolve the issue and allow your Bayesian optimization to run properly.

Follow me so you can message me anytime with future MATLAB questions. If this helps, please accept the answer as well.

[Walter Roberson]

?? The only difference I can find between those two code blocks is in the spacing after commas ? Spaces after commas are good for human readability but make no difference to MATLAB's interpretation.

[Jack]

You're right.

Fix

Inside myfunction, add:

params = table2struct(params);

So your function becomes:

function yfit = myfunction(params, predictors, response, test)
    params = table2struct(params); % Convert params to struct

    Mdl1 = TreeBagger(30, predictors, response, ...
        'Method', "regression", 'Surrogate', "on", ...
        'PredictorSelection', "curvature", ...
        'OOBPredictorImportance', "on", 'MinLeafSize', params.minLS, ...
        'NumPredictorsToSample', params.numPTS);

    yfit = predict(Mdl1, test);
end

This makes sure MATLAB recognizes params.minLS and params.numPTS properly.