How to use PMKL
Written on July 20th, 2021 by control lab
SVM and optimization
The main class of SVM is PMKL. It requires several parameters for Kernel computing and C-SVC or e-SVR.
class PMKLpy.PMKL.PMKL(C = 10, degree = 1, bound = 0.1, epsilon = 0.1, maxit = 100, tol = 1.e-6, probability = False, to_print= True)
Parameters:
- C, float, default=10 \Regularization parameter, smaller values lead to mappings that are more general.
- degree, int, default=1
Degree of the TK Kernel function - bound, float, default=0.1 \Area of integration is [-bound,1+bound]
- epsilon, float, default=0.1 Epsilon parameter for regression problems
- maxit, float default=100 \Maximum number of iterations. Each Iteration includes the SVM training
- tol, float, default=1.e-6 \Tolerance of the TKL optimization algorithm
- probability, boolean, default= False
If True the algorithm are able to predict probability for binary classification. If False the standard SVM problem will be solved - to_print, boolean, default=True
If True the the algorithm will be print the objective function for each iteration, else there will not be any outputs
Attributes:
- fit(X, y) Fit the SVM model according to the given training data.
- predict(X) Perform classification or regression on samples in X.
- predict_proba Compute probabilities of possible outcomes for samples in X (only for classification). IT IS NOT IMPLEMENTED
- get_params IT IS NOT IMPLEMENTED
from PMKLpy import PMKL
SVM = PMKL.PMKL()
SVM.fit(x, y)
ypred = SVM.predict(x)
Feel free to share!
