KNN & PatternList & Categories: To FeatureWeights...

Wrap the selected KNN and use its classification accuracy on the test set constituted by the PatternList and Categories objects as feedback to guide the search for the optimal feature weights. A FeatureWeights object will be created.


Learning rate
The rate at which the maximum distance between the pivot and a random seed is decremented.
Number of seeds
The size of the feature weight neighbourhood.
Stop at
A value specifying a stopping criterion. When feature weights yielding accuracy estimates higher than the specified value the search will stop. A value of 1 imposes no constraints whereas a value of 0.5 will result in the termination of the search algorithm once feature weights resulting in an classification accuracy of 50 percent or better are found.
Specifies whether to search for all features simultaneously or one at a time.
k neighbours
The size of the neighbourhood used for feedback classification.
Vote weighting
The type of vote weighting to be used.

See also:

kNN classifiers Wrapper-based feature weighting
kNN classifiers 1.1.1. Feature weighting
kNN classifiers 1.1. Improving classification accuracy
KNN & PatternList & Categories & FeatureWeights: Evaluate...
kNN classifiers 1. What is a kNN classifier?
kNN classifiers

Links to this page

© Ola Söder, August 9, 2008