Feature Selection

Recursive Feature Elimination

mlpy.rfe_w2(x, y, p, classifier)

RFE algorithm, where the ranking criteria is w^2, described in [Guyon02]. classifier must be an linear classifier with learn() and w() methods.

[Guyon02]

I Guyon, J Weston, S Barnhill and V Vapnik. Gene Selection for Cancer Classification using Support Vector Machines. Machine Learning, 2002.

Parameters :

x: 2d array_like object (N,P)

training data

y : 1d array_like object integer (N)

class labels (only two classes)

p : float [0.0, 1.0]

percentage of features (upper rounded) to remove at each iteration (p=0 one variable)

classifier : object with learn() and w() methods

object

Returns :

ranking : 1d numpy array int

feature ranking. ranking[i] contains the feature index ranked in i-th position.

mlpy.rfe_kfda(x, y, p, lmb, kernel)

KFDA-RFE algorithm based on the Rayleigh coefficient proposed in [Louw06]. The algorithm works with only two classes.

[Louw06]

N Louw and S J Steel. Variable selection in kernel Fisher discriminant analysis by means of recursive feature elimination. Journal Computational Statistics & Data Analysis, 2006.

Parameters :

x: 2d array_like object (N,P)

training data

y : 1d array_like object integer (N)

class labels (only two classes)

p : float [0.0, 1.0]

percentage of features (upper rounded) to remove at each iteration (p=0 one variable)

lmb : float (>= 0.0)

regularization parameter

kernel : mlpy.Kernel object.

kernel.

Returns :

ranking : 1d numpy array int

feature ranking. ranking[i] contains the feature index ranked in i-th position.