How to change the default range of hyperparameters in fitcecoc function?

8 visualizaciones (últimos 30 días)
Is it possible to change the default paramater search range of fitcecoc function? I am trying to find the optimal paramters for SVM in custom range to reduce computational time. For example, I am trying to set below range for following parameters.
BoxConstraint = Positive values log-scaled in the range [1e-3,10]
KernelScale = Positive values log-scaled in the range [1e-3,10]
KernelFunction ='gaussian', , and 'polynomial'
Any suggestions in this regard would be highly appreciated.
Demo Example:
clc
clear all
load fisheriris
t = templateSVM();
results = fitcecoc(meas, species,'Learners',t,'OptimizeHyperparameters','auto',...
'HyperparameterOptimizationOptions',struct('Optimizer',...
'randomsearch'))
T=results.HyperparameterOptimizationResults
  3 comentarios
Don Mathis
Don Mathis el 26 de Jul. de 2021
Walter Roberson's answer is correct. And then you pass 'params' as the value of the 'OptimizeHyperparameters' parameter.
Machine Learning Enthusiast
Machine Learning Enthusiast el 26 de Jul. de 2021
Thank you. How to change other parameters? For instance, for kernal functiion, I am trying to search only guassian and polynomial in the search space.
Name: 'KernelFunction'
Range: {'gaussian' 'linear' 'polynomial'}
Type: 'categorical'
Transform: 'none'
Optimize: 0
Name: 'Coding'
Range: {'onevsall' 'onevsone'}
Type: 'categorical'
Transform: 'none'
Optimize: 1

Iniciar sesión para comentar.

Respuesta aceptada

Walter Roberson
Walter Roberson el 26 de Jul. de 2021
Set nondefault parameters by passing a vector of optimizableVariable objects that have nondefault values. For example,
load fisheriris
params = hyperparameters('fitcecoc',meas,species,'svm');
params(2).Range = [1e-4,1e6];
After you created params look at
{params.Name}
to see which variable is which, to know which one to set the Range for.
  2 comentarios
Machine Learning Enthusiast
Machine Learning Enthusiast el 26 de Jul. de 2021
Thank you. How to change other parameters? For instance, for kernal functiion, I am trying to search only guassian and polynomial in the search space.
Name: 'KernelFunction'
Range: {'gaussian' 'linear' 'polynomial'}
Type: 'categorical'
Transform: 'none'
Optimize: 0
Name: 'Coding'
Range: {'onevsall' 'onevsone'}
Type: 'categorical'
Transform: 'none'
Optimize: 1
Walter Roberson
Walter Roberson el 1 de Ag. de 2021
I think you should be able to do
params(4).Range = {'gaussian', 'polynomial'};

Iniciar sesión para comentar.

Más respuestas (0)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by