Parameter Optimization for categorical parameters?

[HFulcher]

How would I go about creating an optimization loop that chooses different categorical parameters? e.g. finding optimal kernel type in an SVM. It seems like you can only create parameters to search that are numerical.

Thanks

[ipazin]

Hi there @HFulcher,

welcome to KNIME Community!

There is a ticket for this and will add +1 on it for you. Also check this topic where it was discussed:

Feature request: parameter optimisation with Strings

Hi Knimers, I wonder if you can add String parameters to the optimization loop. For Example, During Random Forest parameter Optimisation, I would also like to give Split criteria to the mix and only way to do that is to use proxy variables which makes it more confusing. ( Like 1 = Gini, 2= InformationGain …etc) Being able to define String parameters would definitely ease the pain. Keep up the good work my dear friends.

Br,
Ivan

[HFulcher]

Great, thanks very much @ipazin! Looking forward to seeing this get put into KNIME

[AlexanderFillbrunn]

Hi @HFulcher,
if you want to do grid search, you can also define your parameters using multiple Table Creators, then cross join the tables and use a Table Row To Variable Loop Start for your optimization loop. Just collect the metric you want to maximize in the Loop End and then choose the set of parameters that has the highest score. It does not give you all the other nice optimization options that the Parameter Optimization Loop has, but works just as well for grid search.
Kind regards
Alexander

[HFulcher]

Hi @AlexanderFillbrunn, that’s really helpful, thanks!