varsvm 1.3

Dependencies

varsvm requires:

• Python

• NumPy

• Pandas

• Sklearn

User installation

Install Variant-SVMs using pip

pip install varsvm

or

pip install git+https://github.com/statmlben/varsvm.git

Please install python3-dev before install varsvm

sudo apt-get install python3-dev

Source code

You can check the latest sources with the command:

git clone https://github.com/statmlben/varsvm.git

weightsvm

Classical weighted SVMs.

• class VarSVM.weightsvm(alpha=[], beta=[], C=1., max_iter = 1000, eps = 1e-4, print_step = 1)

  • Parameters:

    • alpha: Dual variable.

    • beta: Primal variable, or coefficients of the support vector in the decision function.

    • C: Penalty parameter C of the error term.

    • max_iter: Hard limit on iterations for coordinate descent.

    • eps: Tolerance for stopping criterion based on the relative l1 norm for difference of beta and beta_old.

    • print_step: If print the interations for coordinate descent, 1 indicates YES, 0 indicates NO.

  • Methods:

    • decision_function(X): Evaluates the decision function for the samples in X.

      • X : array-like, shape (n_samples, n_features)

    • fit(X, y, sample_weight=1.): Fit the SVM model.

      • X : {array-like, sparse matrix}, shape (n_samples, n_features)

      • y : array-like, shape (n_samples,) NOTE: y must be +1 or -1!

      • sample_weight : array-like, shape (n_samples,), weight for each sample.

Drift SVM

SVM with dift or fixed intercept for each instance.

• class VarSVM.driftsvm(alpha=[], beta=[], C=1., max_iter = 1000, eps = 1e-4, print_step = 1)

  • Parameters:

    • alpha: Dual variable.

    • beta: Primal variable, or coefficients of the support vector in the decision function.

    • C: Penalty parameter C of the error term.

    • max_iter: Hard limit on iterations for coordinate descent.

    • eps: Tolerance for stopping criterion based on the relative l1 norm for difference of beta and beta_old.

    • print_step: If print the interations for coordinate descent, 1 indicates YES, 0 indicates NO.

  • Methods:

    • decision_function(X): Evaluates the decision function for the samples in X.

      • X : array-like, shape (n_samples, n_features)

    • fit(X, y, drift, sample_weight=1.): Fit the SVM model.

      • X : {array-like, sparse matrix}, shape (n_samples, n_features)

      • y : array-like, shape (n_samples,). NOTE: y must be +1 or -1!

      • drift: array-like, shape (n_samples,), drift or fixed intercept for each instance, see doc.

      • sample_weight : array-like, shape (n_samples,), weight for each instance.

Non-negative Drift SVM

SVM with non-negative constrains for coefficients.

• class VarSVM.noneg_driftsvm(alpha=[], beta=[], C=1., max_iter = 1000, eps = 1e-4, print_step = 1)

  • Parameters:

    • alpha: Dual variable.

    • beta: Primal variable, or coefficients of the support vector in the decision function.

    • C: Penalty parameter C of the error term.

    • max_iter: Hard limit on iterations for coordinate descent.

    • eps: Tolerance for stopping criterion based on the relative l1 norm for difference of beta and beta_old.

    • print_step: If print the interations for coordinate descent, 1 indicates YES, 0 indicates NO.

  • Methods:

    • decision_function(X): Evaluates the decision function for the samples in X.

      • X : array-like, shape (n_samples, n_features)

    • fit(X, y, drift, sample_weight=1.): Fit the SVM model.

      • X : {array-like, sparse matrix}, shape (n_samples, n_features)

      • y : array-like, shape (n_samples,). NOTE: y must be +1 or -1!

      • drift: array-like, shape (n_samples,), drift or fixed intercept for each instance, see doc.

      • sample_weight : array-like, shape (n_samples,), weight for each instance.

Example

import numpy as np
from sklearn.datasets import make_classification
from varsvm import noneg_driftsvm
from sklearn.model_selection import GridSearchCV

X, y = make_classification(n_features=4, random_state=0)
y = y * 2 - 1

# fit a single model
n = len(X)
drift = .28*np.ones(n)

clf = noneg_driftsvm()
clf.fit(X=X, y=y, drift=drift)
y_pred = clf.decision_function(X=X, drift=drift)

# Tuning hyperparams based on sklearn.model_selection.GridSearchCV
parameters = {'C':[1, 10]}
psvm = noneg_driftsvm()
clf = GridSearchCV(psvm, parameters)
clf.fit(iris.data, iris.target)