Getting Predicted Probabilites from an SVM model in Sklearn

2021-02-06

Intro

SVM is a very diverse model. The general idea is to build a hyperplane that maximizes the margins between classifications. We will cover the the theory in a separate post, but you can think of a building a line in 2-d and a sheet of paper in 3-d.

Building the Linear SVM Model

There are a few different SVM models in Sklearn. We will start by using the simple LinearSVC class. SVM also requires some tuning params. For now, we just specified a default C (cost) of 1.0.

from sklearn.svm import LinearSVC
from sklearn import datasets

iris = datasets.load_iris()
features = iris.data[:100,:2]
target = iris.target[:100]

svc = LinearSVC(C=1.0)

model = svc.fit(features, target)
print(model.score())

Example in 2d

To get a better idea of what is happening, let's plot some a 2d version of our SVM to see what it is creating.

# Load library
from matplotlib import pyplot as plt

# Plot data points and color using their class
color = ["black" if c == 0 else "lightgrey" for c in target]
plt.scatter(features_standardized[:,0], features_standardized[:,1], c=color)

# Create the hyperplane
w = svc.coef_[0]
a = -w[0] / w[1]
xx = np.linspace(-2.5, 2.5)
yy = a * xx - (svc.intercept_[0]) / w[1]

# Plot the hyperplane
plt.plot(xx, yy)
plt.axis("off"), plt.show();

Here, you can see that we create a line through the two classes of observations. Classification won't always be this easy, but you can get a good idea of the intention.

Preprocessing

In the example above, we left out preprocessing in case someone is just looking for the example code. However, it is usually good to at least standardized your data, espceially for SVM. Here is the example again with standardization.

# Load libraries
from sklearn.svm import LinearSVC
from sklearn import datasets
from sklearn.preprocessing import StandardScaler
import numpy as np

# Load data with only two classes and two features
iris = datasets.load_iris()
features = iris.data[:100,:2]
target = iris.target[:100]

# Standardize features
scaler = StandardScaler()
features_standardized = scaler.fit_transform(features)

# Create support vector classifier
svc = LinearSVC(C=1.0)

# Train model
model = svc.fit(features_standardized, target)
GoTea - KoalaTea