from math import *
import random

H = 6
W = []
MW = []
B = []
for i in range(H):
    W.append([random.uniform(-0.5,0.5),random.uniform(-0.5,0.5)])
    MW.append([0,0])
    B.append(random.uniform(-0.5,0.5))

O = []
MO = []
b = 0
for i in range(H):
    O.append(random.uniform(-0.5,0.5))
    MO.append(0)
b = random.uniform(-0.5,0.5)

def sig(x):
    return 1 / (1 + exp(-x))

def sig_prime(x): # x already sigmoided
    return x * (1 - x)

def tanh_prime(x): # x already tanhed
    return 1 - x * x

def forward(i1,i2):
    global W,B,O,b
    h = []
    for i in range(len(W)):
        h.append(sig(W[i][0] * i1 + W[i][1] * i2 + B[i]))
    o = 0
    for i in range(len(O)):
        o += O[i] * h[i]
    o += b
    return sig(o)

def learn(i1,i2,t):
    global W,B,O,b,MW,MO
    alpha = 0.1
    mom = 0.8
    h = []
    for i in range(len(W)):
        h.append(sig(W[i][0] * i1 + W[i][1] * i2 + B[i]))
    o = 0
    for i in range(len(O)):
        o += O[i] * h[i]
    o += b
    o = sig(o)

    e = t - o
    do = e * sig_prime(o)

    eh = []
    for i in range(len(O)):
        eh.append(do * O[i])

    dh = []
    for i in range(len(O)):
        dh.append(eh[i] * sig_prime(h[i]))

    for i in range(len(O)):
        MO[i] = mom * MO[i] + alpha * h[i] * do
        O[i] += MO[i]
    b += alpha * do

    for i in range(len(W)):
        MW[i][0] = mom * MW[i][0] + alpha * i1 * dh[i]
        MW[i][1] = mom * MW[i][1] + alpha * i2 * dh[i]
        W[i][0] += MW[i][0]
        W[i][1] += MW[i][1]
        B[i] += alpha * dh[i]

inputs = [[0,0],[0,1],[1,0],[1,1]]
targets = [0,1,1,0]

for n in range(10000):
    a = list(range(len(inputs)))
    random.shuffle(a)
    for i in a:
        learn(inputs[i][0],inputs[i][1],targets[i])

    if (n+1) % 1000 == 0:
        cost = 0
        for i in a:
            x = forward(inputs[i][0],inputs[i][1])
            cost += (targets[i]-x) * (targets[i]-x)
        print("cost ", cost/4)

for i in range(len(inputs)):
    print(inputs[i][0],inputs[i][1],forward(inputs[i][0],inputs[i][1]))