This is one of the 100 recipes of the IPython Cookbook, the definitive guide to high-performance scientific computing and data science in Python.

8.5. Using Support Vector Machines for classification tasks

  1. Let's do the traditional imports.
In [ ]:
import numpy as np
import pandas as pd
import sklearn
import sklearn.datasets as ds
import sklearn.cross_validation as cv
import sklearn.grid_search as gs
import sklearn.svm as svm
import matplotlib as mpl
import matplotlib.pyplot as plt
%matplotlib inline
  1. We generate 2D points and assign a binary label according to a linear operation on the coordinates.
In [ ]:
X = np.random.randn(200, 2)
y = X[:, 0] + X[:, 1] > 1
  1. We now fit a linear Support Vector Classifier (SVC). This classifier tries to separate the two groups of points with a linear boundary (a line here, more generally a hyperplane).
In [ ]:
# We train the classifier.
est = svm.LinearSVC(), y);
  1. We define a function that displays the boundaries and decision function of a trained classifier.
In [ ]:
# We generate a grid in the square [-3,3 ]^2.
xx, yy = np.meshgrid(np.linspace(-3, 3, 500),
                     np.linspace(-3, 3, 500))
# This function takes a SVM estimator as input.
def plot_decision_function(est):
    # We evaluate the decision function on the grid.
    Z = est.decision_function(np.c_[xx.ravel(), yy.ravel()])
    Z = Z.reshape(xx.shape)
    cmap =
    # We display the decision function on the grid.
                extent=(xx.min(), xx.max(), yy.min(), yy.max()),
                aspect='auto', origin='lower', cmap=cmap);
    # We display the boundaries.
    plt.contour(xx, yy, Z, levels=[0], linewidths=2,
    # We display the points with their true labels.
    plt.scatter(X[:, 0], X[:, 1], s=30, c=.5+.5*y, lw=1, 
                cmap=cmap, vmin=0, vmax=1);
    plt.axhline(0, color='k', ls='--');
    plt.axvline(0, color='k', ls='--');
    plt.axis([-3, 3, -3, 3]);
  1. Let's take a look at the classification results with the linear SVC.
In [ ]:
plt.title("Linearly separable, linear SVC");

The linear SVC tried to separate the points with a line and it did a pretty good job.

  1. We now modify the labels with a XOR function. A point's label is 1 if the coordinates have different signs. This classification is not linearly separable. Therefore, a linear SVC fails completely.
In [ ]:
y = np.logical_xor(X[:, 0] > 0, X[:, 1] > 0)
# We train the classifier.
est = gs.GridSearchCV(svm.LinearSVC(), 
                      {'C': np.logspace(-3., 3., 10)});, y);
print("Score: {0:.1f}".format(
      cv.cross_val_score(est, X, y).mean()))
# Plot the decision function.
plt.title("XOR, linear SVC");
  1. Fortunately, it is possible to use non-linear SVCs by using non-linear kernels. Kernels specify a non-linear transformation of the points into a higher-dimensional space. Transformed points in this space are assumed to be more linearly separable, although they are not necessarily in the original space. By default, the SVC classifier in scikit-learn uses the Radial Basis Function (RBF) kernel.
In [ ]:
y = np.logical_xor(X[:, 0] > 0, X[:, 1] > 0)
est = gs.GridSearchCV(svm.SVC(), 
                      {'C': np.logspace(-3., 3., 10),
                    'gamma': np.logspace(-3., 3., 10)});, y);
print("Score: {0:.3f}".format(
      cv.cross_val_score(est, X, y).mean()))
plt.title("XOR, non-linear SVC");

This time, the non-linear SVC does a pretty good job at classifying these non-linearly separable points.

You'll find all the explanations, figures, references, and much more in the book (to be released later this summer).

IPython Cookbook, by Cyrille Rossant, Packt Publishing, 2014 (500 pages).