import numpy as np
err = np.array([])
print (err)

def sigm(x):
	return (1 / (np.exp(-x) + 1))

train_in = np.array ([ [1, 0, 1, 0],
									[1, 1, 1, 1],
									[0, 0, 0, 0],
									[1, 1, 1, 1],
									[0, 1, 0, 1],
									[0, 0, 0, 1],
									[0, 1, 0, 0],
									[1, 0, 0, 1],
									[1, 0, 0, 1] ])

train_out = np.array ([1, 1, 0, 1, 0, 0, 0, 1, 1])

np.random.seed(1)
weight = 2 * np.random.random((4, 1)) - 1
#print(weight)

for i in range (20000):
	input = train_in
	output = sigm(np.dot(input,  weight))
	print("out ", output)

	for a in output.shape:
	#
		err = np.add(train_out[a] - output[a])
		print(err)

	d_weight = np.dot(train_in.T, err * (output * (1- output)))
	print("weight ", d_weight)
	weight += d_weight

#print(output)