linear_classifier.py

import numpy as np
import matplotlib.pyplot as plt
from sklearn.preprocessing import OneHotEncoder
from sklearn.cross_validation import train_test_split

np.random.seed(3)

center0 = 3 * np.random.randint(-2, 2, 2)
center1 = 3 * np.random.randint(-2, 2, 2)

n_samples = 3000
X = np.zeros((n_samples, 3))
X[:, 0] = np.tile(1, n_samples)
X[:n_samples//2, 1:] = center0 + 1.5 * np.random.randn(n_samples//2, 2)
X[n_samples//2:, 1:] = center1 + 1.5 * np.random.randn(n_samples//2, 2)

t = [0] * (n_samples//2) + [1] * (n_samples//2)
t = np.array(t)[:, np.newaxis]

enc = OneHotEncoder()
T = enc.fit_transform(t).toarray()

X_train, X_test, T_train, T_test = train_test_split(X, T)

def linear_weight(X, T):
    return np.linalg.solve(np.dot(X.T, X), np.dot(X.T, T))

W_mse = linear_weight(X_train, T_train)

Y_test = np.dot(X_test, W_mse)

# accuracy
acc = np.mean(np.argmax(Y_test, axis=1) == np.argmax(T_test, axis=1))
print(acc)

# plot
label0 = T_test[:, 0] == 1
label1 = T_test[:, 1] == 1
plt.scatter(X_test[label0, 1], X_test[label0, 2], c='b')
plt.scatter(X_test[label1, 1], X_test[label1, 2], c='r')

# y = w0 + w1 x
w0 = (W_mse[0, 1] - W_mse[0, 0]) / (W_mse[2, 0] - W_mse[2, 1])
w1 = (W_mse[1, 1] - W_mse[1, 0]) / (W_mse[2, 0] - W_mse[2, 1])
x = np.linspace(X_test[:, 1].min(), X_test[:, 1].max(), 1000)
plt.plot(x, w0+w1*x)
plt.show()