How to select an optimal parameter set

Previous  Top  Next

The tuning results for the various parameter sets are displayed in a list view on the tab sheet Tuning Results. The first part of the columns contains the values of the parameter sets and the second part contains the corresponding model characteristics and the prediction accuracy.

Click on the header of the column Error to sort the parameter sets with respect to the corresponding prediction accuracy. Select the best parameter set as long as it is not colored red or yellow (light yellow is ok) by double clicking it. Alternatively, you may also highlighting it and hit the button Select.


Why are some parameter sets colored red or yellow ?
Colored parameter sets contain unfavorable parameter values. A set is colored red, if pruning is disabled and it is colored yellow, if the corresponding model size approaches the maximum size thresholds.


Some additional rules of thumb

·

Prefer moderate parameter values instead of extreme ones

·

Be careful, if the prediction accuracy is the same for several parameter sets. Maybe you should repeat the tuning then. Also it should help to increase the number of cross-validations or repetitions.

·

Chose the 3 or 4 best parameter sets (note down their ID) and repeat the tuning. Do you get similar results? Are the parameter sets, that were the best in the previous run, again within the best sets?