A=FISHER(H) returns a fisher object initialized with hyperparameters H.
Calculates a Fisher/Correlation score for each feature to implement
feature selection.
Hyperparameters, and their defaults
feat=[] -- number of features
output_rank=1 -- set to 1 if only the feature ranking matters
(does not perform any classification on the data)
otherwise performs classification using
weights given by individual correlation scores
method=2 -- useful only for multi-class. Set the how to combine
the score of different one-vs-rest fisher's score.
(2 = take the sum, 1 = take the max)
Model
w -- the weights
b0 -- the threshold (when using all features)
rank -- the ranking of the features
d -- training set
Methods:
train, test, get_w
Example:
d=gen(toy); a=fisher; a.feat=10; a.output_rank=1;[r,a]=train(a,d);
a.rank - lists the chosen features in order of importance
Note:
To use for Furey et al. method (e.g correlation coefficients + svm)
use: chain(fisher('output_rank=1'),svm)
|
Reference : Neural Networks for Pattern Recognition |