simple_perceptron.py

import numpy as np
import matplotlib.pyplot as plt

# prepare data

d = 2
N = 3
mean = 5

rng = np.random.RandomState(100)
x1 = rng.randn(N, d)
x2 = rng.randn(N, d) + np.array([mean, mean])
x = np.concatenate((x1, x2), axis=0)

# plot data
# plt.scatter(x[:,0], x[:,1])
# plt.show()

# init weight and bias
w = np.zeros(d)
b = 0

def y(x):
  return step(np.dot(x, w) + b)

def step(x):
  return 1 * (x > 0)

def t(x):
  if i < N:
    return 0
  else:
    return 1

while True:
  classified = True
  for i in range(N * 2):
    delta_w = (t(i) - y(x[i])) * x[i]
    delta_b = (t[i] - y(x[i]))
    w += delta_w
    b += delta_b
    classified *= all(delta_w == 0) * (delta_b == 0)
  if classified:
    break

print(weight, b)