Parameters of the PNC2 algorithm
Previous  Top  Next

The parameters used to learn a model are located on the bottom of the tab sheet Basic.


Parameter
Explanation
note
Intervals


Number of bins used to discretize a continuous output

Note: Only valid for regression tasks More Info


note
Anti COD


Reduce the curse of dimensionality (COD) effect in high dimensional input spaces


note
Eta


Sets a threshold for the hitrate of a rule to pass the merge test. A hitrate of 100% is required, if this value is set to 0. If you chose an value of 1, then the required hitrate is 50%.


note
Kernel width


Controls the k-nearest-neighbor component of the prediction mechanism for classification tasks. Set it to 1, to follow a pure 1-nearest-rule strategy.


note
Sigma


Controls the k-nearest-neighbor component of the prediction mechanism for regression tasks


note
Min. rule mass


Threshold for the mass of a rule. All rules with a lower mass are reduced. The mass of a rule is the number of learn data tuples merged within the equivalent cluster. Rules with a low mass have weak statistical support. More Info


note
Kill small rules  


If you enable this option, then all rules with a mass less than Min. rule mass are deleted permanently from the model. Thus, later on when using the model, it will have no effect if you chose a smaller value for Min. rule mass.


note
Prune rules


Enable the context-sensitive feature selection which individually eliminates irrelevant inputs from each rule. You should always enable this option!


note
Prune anyway


If you enable this option, then the calculations necessary for the context-sensitive feature selection are executed even if pruning is disabled, to allow you, to enable this option later on when using the model.


note
Use Weights


Use feature weights calculated on the basis of the transinformation of the particular input to the output


note
Euclidean
distance

Use Euclidean or Manhattan City-Block distance



Hint: It is always a good idea to use the default values. You have to press the Defaults button after changing the output column to re-calculate appropriate values.  

Note: The parameters can only be edited if the check box
 Skip Tuning is checked.


More Info