eps = 1.
elif algorithm == 'logistic_regression':
eps = 5.
elif algorithm == 'polynomial_regression':
X = polynomial_transform(X)
eps = 100
nSamples = X.shape[0]
plt.ion()
w = np.zeros(X.shape[1])
b = 0.
for i in range(100):
if algorithm == 'perceptron':
# Predicted outputs
T = (np.dot(X, w) + b) > 0
C = T
else:
# Predicted outputs
T = sigm(np.dot(X, w) + b)
C = (T > .5)
w += eps*np.dot(Y - T, X)/nSamples
b += eps*np.mean(Y - T)
# Draw the decision boundary.
plt.clf()
plt.title('p = ' + str(X.shape[1]) + ', Iteration = ' + str(i) + ', Error = ' + str(np.mean(Y != C)))
decision_boundary(Xorig, Y, w, b)
import numpy as np
import matplotlib.pyplot as plt
from decision_boundary import *
# Load the dataset
with np.load('./data.npz') as data:
X = data['X']
Y = data['Y']
w = np.zeros(X.shape[1])
b = 0.
nSamples = X.shape[0]
plt.ion()
for i in range(100):
# Predicted outputs
T = np.sign(np.dot(X, w) + b)
w += .01 * np.dot(Y - T, X) / nSamples
b += .01 * np.mean(Y - T)
# Draw the decision boundary.
plt.clf()
plt.scatter(X[:, 0], X[:, 1], c=Y, cmap='winter')
plt.axis([-2, 4, -5, 3])
plt.title('Iteration = ' + str(i) + ', Error = ' + str(np.mean(Y != T)))
plt.axis('off')
decision_boundary(w, b)
w = np.zeros(X.shape[1])
b = 0.
T = np.zeros(nSamples)
C = np.zeros(nSamples)
sum_gradients = np.zeros(X.shape[1])
gradients = np.zeros(nSamples)
for i in range(100 * nSamples):
# Draw an example at random
n = np.random.randint(nSamples)
# Predicted outputs
T[n] = sigm(np.dot(X[n], w) + b)
C[n] = (T[n] > .5)
sum_gradients -= gradients[n] * X[n]
gradients[n] = Y[n] - T[n]
sum_gradients += gradients[n] * X[n]
w += eps * sum_gradients / nSamples
b += eps * np.mean(gradients[n])
if i % nSamples == 0:
# Draw the decision boundary.
plt.clf()
plt.title('p = ' + str(X.shape[1]) + ', Iteration = ' +
str(i / nSamples) + ', Error = ' + str(np.mean(Y != C)))
decision_boundary(Xorig, Y, w, b)
import numpy as np
import matplotlib.pyplot as plt
from decision_boundary import *
# Load the dataset
with np.load('./data.npz') as data:
X = data['X']
Y = data['Y']
w = np.zeros(X.shape[1])
b = 0.
nSamples = X.shape[0]
plt.ion()
for i in range(100):
# Predicted outputs
T = np.sign(np.dot(X, w) + b)
w += .01*np.dot(Y - T, X)/nSamples
b += .01*np.mean(Y - T)
# Draw the decision boundary.
plt.clf()
plt.scatter(X[:, 0], X[:, 1], c=Y, cmap='winter')
plt.axis([-2, 4, -5, 3])
plt.title('Iteration = ' + str(i) + ', Error = ' + str(np.mean(Y != T)))
plt.axis('off')
decision_boundary(w, b)
