PRTools contents |
FEATEVAL
J = FEATEVAL(A,CRIT,T)
J = FEATEVAL(A,CRIT,N)
J = A*FEATEVAL([],CRIT,T)
J = A*FEATEVAL([],CRIT,N)
Input | |
A | input dataset |
CRIT | string name of a method or untrained mapping, default 'NN' |
T | validation dataset (optional) |
N | number of cross-validations (optional) |
Output | |
J | scalar criterion value |
Evaluation of features by the criterion CRIT, using objects in the dataset A. The larger J, the better. Resulting J-values are incomparable over the various methods. The following methods are supported
crit='in-in' : inter-intra distance.
crit='maha-s': sum of estimated Mahalanobis distances.
crit='maha-m': minimum of estimated Mahalanobis distances.
crit='eucl-s': sum of squared Euclidean distances.
crit='eucl-m': minimum of squared Euclidean distances.
crit='NN' : 1-Nearest Neighbour leave-one-out
classification performance (default).
(performance = 1 - error).
crit='mad' : mean absolute deviation (only for regression!)
crit='mse' : mean squared error (only for regression!)
For classification problems, CRIT can also be any untrained
classifier, e.g. LDC([],1e-6,1e-6). Then the classification error is
used for a performance estimate. If supplied, the dataset T is used
for obtaining an unbiased estimate of the performance of classifiers
trained with the dataset A. If a number of cross-validations N is
supplied, the routine is run for N times with different training and
test sets generated from A by cross-validation. Results are averaged.
If T nor N are given, the apparent performance on A is used.
datasets, featselo, featselb, featself, featselp, featselm, featrank,
PRTools contents |