How to get optimal tree when using random forest method

Hi,

I use the 

    meas=[...   
       [53.4
        194.5
        220.5
        273.1
        312.3
        331.7
        159.7
        199.5
        293.9
        390.7
        392.1
        398.9
        399.1
        400.3
        52.1
        249.3
        357.7
        44.8
        163.9
        200.5
        335.1
        92.4
        130.3
        77.1]]

    species=[{'v16'
        'v6'
        'v1'
        'v126'
        'v6'
        'v1'
        'v16'
        'v6'
        'v5'
        'v156'
        'v6'
        'v1'
        'v16'
        'v6'
        'v12456'
        'v56'
        'v256'
        'v2456'
        'v6'
        'v16'
        'v1234'
        'v16'
        'v12356'
        'v6'}]

I use below code:

    b = TreeBagger(50,meas,species,'oobpred','on')

But how to get optimal tree? Many thanks in advance.

Answer 1

[mizuki]

Optimize tree with Bayesian Optimization (use bayesopt function). This feature is introduced in R2016b.

• Tune Random Forest Using Quantile Error and Bayesian Optimization

https://jp.mathworks.com/help/stats/tune-random-forest-using-quantile-error-and-bayesian-optimization.html

Also you can search optimal parameters with other methods such as gridsearch, but you need to write code. For gridsearch, this might help you:

• How to do grid Search to optimize sigma using Matlab?

https://jp.mathworks.com/matlabcentral/answers/225681-how-to-do-grid-search-to-optimize-sigma-using-matlab

[Amy Xu]

I'm trying to run the following Mathworks example with my own X and Y:

"Tune Random Forest Using Quantile Error and Bayesian Optimization"

https://jp.mathworks.com/help/stats/tune-random-forest-using-quantile-error-and-bayesian-optimization.html

But, I'm getting the following error:

Undefined function or variable 'Y'.

I have attached the modified code (place both files in one folder on your PC drive). Can anyone help?

https://www.mathworks.com/matlabcentral/answers/336178-how-to-use-bayesian-optimization

[Marta Caneda Portela]

What if I also need 5-fold cross-validation? I can't find any examples online

Answer 2

[Don Mathis]

You could also try to find the best ensemble like this:

fitcensemble(meas,species,'OptimizeHyperparameters','all')

This will also try boosted decision trees in addition to random forests.