Feature Weighting/Selection Sun08
A feature weighting/selection algorithm described in [Sun08].
-
class mlpy.FSSun(T=1000, sigma=1.0, theta=0.001, lmbd=1.0, eps=0.001, alpha0=1.0, c=0.01, rho=0.5)
Sun Algorithm for feature weighting/selection
Initialize the FSSun class
Parameters: |
- T : int (> 0)
max loops
- sigma : float (> 0.0)
kernel width
- theta : float (> 0.0)
convergence parameter
- lmbd : float
regularization parameter
- eps : float (0 < eps << 1)
termination tolerance for steepest descent method
- alpha0 : float (> 0.0)
initial step length (usually 1.0) for line search
- c : float (0 < c < 1/2)
costant for line search
- rho : flaot (0 < rho < 1)
alpha coefficient for line search
|
New in version 2.0.9.
-
weights(x, y)
Compute the feature weights
Parameters: |
- x : 2d ndarray float (samples x feats)
training data
- y : 1d ndarray integer (-1 or 1)
classes
|
Returns: |
- fw : 1d ndarray float
feature weights
|
Attributes: |
- FSSun.loops : int
number of loops
|
Raises: |
- ValueError
if classes are not -1 or 1
- SigmaError
if sigma parameter is too small
|
[Sun08] | Yijun Sun, S. Todorovic, and S. Goodison. A Feature Selection Algorithm Capable of Handling Extremely Large Data Dimensionality. In Proc. 8th SIAM International Conference on Data Mining (SDM08), pp. 530-540, April 2008. |