深度学习：通过感知器实现and函数

通过梯度下降思想实现感知器：

import random
import numpy as np

# and函数是二元函数 需要两个权值向量
b = random.random()
w1 = random.random()
w2 = random.random()
print(w1, w2, b)
# 学习率
a = 0.1
# 真值表
x = np.array([[1,1], [0,0], [1,0], [0,1]])
t = np.array([1, 0, 0, 0])
eps = 0.01
Z = 0
for i in range(0, 4):
    Z = b + x[i, 0] * w1 + x[i, 1] * w2

def S(x):
    return 1/(1+(2.7)**(-x))

Dw1 = 1
Dw2 = 1
Db  = 1
while Dw1 >= eps or Dw2 >= eps or Db >= eps:
    for i in range (0, 4):
        delta = t[i] - S(Z)
        w1 = w1 + a * delta * x[i, 0]
        w2 = w2 + a * delta * x[i, 1]
        b = b + a * delta
    Z = b + x[i, 0] * w1 + x[i, 1] * w2
    Dw1 = 2 * (t[0] - S(Z)) * S(Z) *(1 - S(Z)) * x[0, 0]
    Dw2 = 2 * (t[0] - S(Z)) * S(Z) *(1 - S(Z)) * x[0, 1]
    Db  = 2 * (t[0] - S(Z)) * S(Z) *(1 - S(Z))

---------------------------------------------------------------------------

KeyboardInterrupt                         Traceback (most recent call last)

<ipython-input-199-308254797349> in <module>
     11     Dw1 = 2 * (t[0] - S(Z)) * S(Z) *(1 - S(Z)) * x[0, 0]
     12     Dw2 = 2 * (t[0] - S(Z)) * S(Z) *(1 - S(Z)) * x[0, 1]
---> 13     Db  = 2 * (t[0] - S(Z)) * S(Z) *(1 - S(Z))


<ipython-input-198-05863795717d> in S(x)
      1 def S(x):
----> 2     return 1/(1+(2.7)**(-x))


KeyboardInterrupt: 

print(w1, w2, b)

12305.25097849656 12305.02886071489 -12305.726717189342

def predict(x1, x2):
    return S(b + w1 * x1 + w2 * x2)

print('1 and 1 = %d' % predict(1, 1))
print('0 and 0 = %d' % predict(0, 0))
print('1 and 0 = %d' % predict(1, 0))
print('0 and 1 = %d' % predict(0, 1))

1 and 1 = 1
0 and 0 = 0
1 and 0 = 0
0 and 1 = 0


<ipython-input-198-05863795717d>:2: RuntimeWarning: overflow encountered in double_scalars
  return 1/(1+(2.7)**(-x))

但是由于eps取合适值较难，所以while循环需要强行跳出。中断后继续执行，and函数实现。

也可以使用for循环来实现：

import random
import numpy as np

# and函数是二元函数 需要两个权值向量
b = random.random()
w1 = random.random()
w2 = random.random()
# 学习率
a = 0.1
# 真值表
x = np.array([[1,1], [0,0], [1,0], [0,1]])
t = np.array([1, 0, 0, 0])
# 
eps = 0.001
# 求和Z
Z = 0
for i in range(0, 4):
    Z = b + x[i, 0] * w1 + x[i, 1] * w2
def S(x):
    return 1/(1+(2.7)**(-x))
for i in range (0,10): # 迭代十次
    for i in range (0, 4):
        Z = b + x[i, 0] * w1 + x[i, 1] * w2
        delta = t[i] - S(Z)
        w1 = w1 + a * delta * x[i, 0]
        w2 = w2 + a * delta * x[i, 1]
        b = b + a * delta

def predict(x1, x2):
    ans = S(b + w1 * x1 + w2 * x2)
    if(ans > 0.5):
        return 1
    else:
        return 0

print('1 and 1 = %d' % predict(1, 1))
print('0 and 0 = %d' % predict(0, 0))
print('1 and 0 = %d' % predict(1, 0))
print('0 and 1 = %d' % predict(0, 1))

1 and 1 = 1
0 and 0 = 0
1 and 0 = 0
0 and 1 = 0