Kernel_trick_idea.svg


Summary

Description
English: An illustration of kernel trick in SVM. Here the kernel is given by:
Date
Source Own work
Author Shiyu Ji

Python Source Code

import numpy as np
import matplotlib
matplotlib.use('svg')
import matplotlib.pyplot as plt
from sklearn import svm
from matplotlib import cm

# Prepare the training set.
# Suppose there is a circle with center at (0, 0) and radius 1.2.
# All the points within the circle are labeled 1.
# All the points outside the circle are labeled 0.
nSamples = 100
spanLen = 2
X = np.zeros((nSamples, 2))
y = np.zeros((nSamples, ))

for i in range(nSamples):
  a, b = [np.random.uniform(-spanLen, spanLen) for _ in ['x', 'y']]
  X[i][0], X[i][1] = a, b
  y[i] = 1 if a*a + b*b < 1.2*1.2 else 0

# Custom kernel,
def my_kernel(A, B):
  gram = np.zeros((A.shape[0], B.shape[0]))
  for i in range(A.shape[0]):
    for j in range(B.shape[0]):
      assert A.shape[1] == B.shape[1]
      L2A, L2B = 0.0, 0.0
      for k in range(A.shape[1]):
        gram[i, j] += A[i, k] * B[j, k]
        L2A += A[i, k] * A[i, k]
        L2B += B[j, k] * B[j, k]
      gram[i, j] += L2A * L2B
  return gram

# SVM train.
clf = svm.SVC(kernel = my_kernel)
clf.fit(X, y)
coef = clf.dual_coef_[0]
sup = clf.support_
b = clf.intercept_
x_min, x_max = -spanLen, spanLen
y_min, y_max = -spanLen, spanLen
xx, yy = np.meshgrid(np.arange(x_min, x_max, .02), np.arange(y_min, y_max, .02))
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
Z = Z.reshape(xx.shape)

# Plot the 2D layout.
fig = plt.figure(figsize = (6, 14))
plt1 = plt.subplot(121)
plt1.set_xlim([-spanLen, spanLen])
plt1.set_ylim([-spanLen, spanLen])
plt1.set_xticks([-1, 0, 1])
plt1.set_yticks([-1, 0, 1])
plt1.pcolormesh(xx, yy, Z, cmap=cm.Paired)
y_unique = np.unique(y)
colors = cm.rainbow(np.linspace(0.0, 1.0, y_unique.size))
for this_y, color in zip(y_unique, colors):
  this_Xx = [X[i][0] for i in range(len(X)) if y[i] == this_y]
  this_Xy = [X[i][1] for i in range(len(X)) if y[i] == this_y]
  plt1.scatter(this_Xx, this_Xy, c=color, alpha=0.5)

# Process the training data into 3D by applying the kernel mapping:
# phi(x, y) = (x, y, x*x + y*y).
X3d = np.ndarray((X.shape[0], 3))
for i in range(X.shape[0]):
    a, b = X[i][0], X[i][1]
    X3d[i, 0], X3d[i, 1], X3d[i, 2] = [a, b, a*a + b*b]

# Plot the 3D layout after applying the kernel mapping.
from mpl_toolkits.mplot3d import Axes3D
plt2 = plt.subplot(122, projection="3d")
plt2.set_xlim([-spanLen, spanLen])
plt2.set_ylim([-spanLen, spanLen])
plt2.set_xticks([-1, 0, 1])
plt2.set_yticks([-1, 0, 1])
plt2.set_zticks([0, 2, 4])
for this_y, color in zip(y_unique, colors):
  this_Xx = [X3d[i, 0] for i in range(len(X3d)) if y[i] == this_y]
  this_Xy = [X3d[i, 1] for i in range(len(X3d)) if y[i] == this_y]
  this_Xz = [X3d[i, 2] for i in range(len(X3d)) if y[i] == this_y]
  plt2.scatter(this_Xx, this_Xy, this_Xz, c=color, alpha=0.5)

# Plot the 3D boundary.
def onBoundary(x, y, z, X3d, coef, sup, b):
  err = 0.0
  n = len(coef)
  for i in range(n):
    err += coef[i] * (x*X3d[sup[i], 0] + y*X3d[sup[i], 1] + z*X3d[sup[i], 2])
  err += b
  if abs(err) < .1:
    return True
  return False

Xr = np.arange(x_min, x_max, .02)
Yr = np.arange(y_min, y_max, .02)
Z = np.zeros(Z.shape)
for i in range(Xr.shape[0]):
  x = Xr[i]
  for j in range(Yr.shape[0]):
    y = Yr[j]
    for z in np.arange(0, 2, .02):
      if onBoundary(x, y, z, X3d, coef, sup, b):
        Z[i, j] = z
        break
plt2.plot_surface(xx, yy, Z, cmap='summer', alpha=0.2)

plt.savefig("kernel_trick_idea.svg", format = "svg")

Licensing

I, the copyright holder of this work, hereby publish it under the following license:
w:en:Creative Commons
attribution share alike
You are free:
  • to share – to copy, distribute and transmit the work
  • to remix – to adapt the work
Under the following conditions:
  • attribution – You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.
  • share alike – If you remix, transform, or build upon the material, you must distribute your contributions under the same or compatible license as the original.

Captions

Add a one-line explanation of what this file represents

Items portrayed in this file

depicts

27 June 2017