Resulted network does not seem to work

Hi, I ran the Neat Python XOR example and a neural network it found has the following properties:

Best individual in generation 22 meets fitness threshold - complexity: (2, 8) Number of evaluations: 3450

Best genome: Nodes: NodeGene(id=0, type=INPUT, bias=0.0, response=4.924273, activation=sigmoid) NodeGene(id=1, type=INPUT, bias=0.0, response=4.924273, activation=sigmoid) NodeGene(id=2, type=OUTPUT, bias=0.0, response=4.921960812206318, activation=sigmoid) NodeGene(id=3, type=HIDDEN, bias=-2.340619398413843, response=4.930091408143276, activation=sigmoid) NodeGene(id=4, type=HIDDEN, bias=0.4674809078722945, response=4.918937894580367, activation=sigmoid) Connections: ConnectionGene(in=0, out=2, weight=1.0534262789427578, enabled=True, innov=0) ConnectionGene(in=1, out=2, weight=-1.6479324848766828, enabled=True, innov=1) ConnectionGene(in=0, out=3, weight=-3.5479647367966765, enabled=True, innov=2) ConnectionGene(in=3, out=2, weight=5.970619032226201, enabled=True, innov=3) ConnectionGene(in=1, out=3, weight=4.634338591600734, enabled=True, innov=4) ConnectionGene(in=0, out=4, weight=0.19203601809123594, enabled=True, innov=5) ConnectionGene(in=4, out=2, weight=-0.6588709650159562, enabled=True, innov=6) ConnectionGene(in=4, out=3, weight=-3.0010419817357072, enabled=True, innov=8) Node order: [4, 3]

Output: expected 0.00000 got 0.11990 expected 1.00000 got 1.00000 expected 1.00000 got 0.92939 expected 0.00000 got 0.00395 {2} {2} {0, 1, 3, 4} {0, 1, 2, 3, 4}

However, when trying to implement the neural network in Python, I don’t get the same results:

import math

def sigmoid(x):
    f = 1 / (1 + math.exp(-x))
    return f


node0 = 1
node1 = 1
node2 = 0

node3_bias = -2.340619398413843
node4_bias = 0.4674809078722945

connection_0_2 = 1.0534262789427578
connection_1_2 = -1.6479324848766828
connection_0_3 = -3.5479647367966765
connection_3_2 = 5.970619032226201
connection_1_3 = 4.634338591600734
connection_0_4 = 0.19203601809123594
connection_4_2 = -0.6588709650159562
connection_4_3 = -3.0010419817357072


def main():
    activated_node4 = sigmoid(node4_bias * 1 +
                              node0 * connection_0_4)
    activated_node3 = sigmoid(node3_bias * 1 +
                              activated_node4 * connection_4_3 +
                              node0 * connection_0_3 +
                              node1 * connection_1_3)
    activated_node2 = sigmoid(activated_node4 * connection_4_2 +
                              activated_node3 * connection_3_2 +
                              node1 * connection_1_2 +
                              node0 * connection_0_2)

    print("The value is: ", activated_node2)


if __name__ == "__main__":

    main()

For nodes 1 and 1 I get:

The value is: 0.30957048732254944

which does not match the expected output above.

Can anyone please help me with some pointers as to what I am getting wrong?

Thank you.

[radu-nedelcu-de]

Hi guys,

Thank you so much! I was ignoring both the response as I did not know it is required and the sigmoid activation function, thinking it would be identical.

It’s working fine now.

[CodeReclaimers]

Glad to help–it’s always those little back-end details that are hard to spot right away. 😃 Thanks for taking an evolved network and doing the experiment of using it elsewhere!