import math

class MLP():
    learning_rate = 0.1
    epoch = 2000
    display = 100
    w11=0.5;w12=0.9;w21=0.4;w22=1.0;w31=-1.2;w32=1.1;b1=-1;b2=-1;b3=-1

    def train(self):
        step = 1
        while(step < self.epoch):
            for i in range(0,4):
                result= self.train_net(int(i/2),i%2)
                if step % self.display==0:
                    print("epoch:",step,"input1:",int(i/2),"input2:",i%2,"output:",result)
            step+=1

    def train_net(self,input1,input2):
        net1 = self.w11*input1+self.w21*input2+self.b1
        net2 = self.w12*input1+self.w22*input2+self.b2
        y1 = self.sigmoid(net1)
        y2 = self.sigmoid(net2)
        net3 = y1*self.w31+y2*self.w32+self.b3
        y3 = self.sigmoid(net3)
        target = int(input1) & int(input2)
        deviation3 = y3*(1-y3)*(target-y3)
        deviation1 = self.w31*deviation3*y1*(1-y1)
        deviation2 = self.w32*deviation3*y2*(1-y2)
        self.w11 += self.learning_rate*deviation1*input1
        self.w12 += self.learning_rate*deviation2*input2
        self.w21 += self.learning_rate*deviation1*input1
        self.w22 += self.learning_rate*deviation2*input2
        self.w31 += self.learning_rate*deviation3*y1
        self.w32 += self.learning_rate*deviation3*y2
        self.b1 += self.learning_rate*deviation1
        self.b2 += self.learning_rate*deviation2
        self.b3 += self.learning_rate*deviation3

#        print("input1 :"+str(input1)+" input2 :"+str(input2))
#        print("output :"+str(y3))
        #print("w11:"+str(self.w11))
        #print("w12:"+str(self.w12))
        #print("w21:"+str(self.w21))
        #print("w22:"+str(self.w22))
        #print("w31:"+str(self.w31))
        #print("w32:"+str(self.w32))
        #print("b1:"+str(self.b1))
        #print("b2:"+str(self.b2))
        #print("b3:"+str(self.b3))
        return y3

    def sigmoid(self, net):
        return (1/(1+math.exp(-net)))

if __name__== '__main__':
    mlp = MLP()
    mlp.train()