Machine Learning Algorithms - Second Edition

Custom kernels

Normally, built-in kernels can efficiently solve most real-life problems; however, scikit-learn allows us to create custom kernels as normal Python functions:

import numpy as np  def custom_kernel(x1, x2):     return np.square(np.dot(x1, x2) + 1)

The function can be passed to SVC through the kernel parameter, which can assume fixed string values ('linear', 'rbf', 'poly', and 'sigmoid') or a callable (such as kernel=custom_kernel).