-
Notifications
You must be signed in to change notification settings - Fork 0
/
NeuralNetwork.py
63 lines (49 loc) · 2.39 KB
/
NeuralNetwork.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
import numpy as np
from NeuralLayer import NeuralLayer as NeuralLayer
class NeuralNetwork:
def __init__(self, topology, activate_function, derivative_activate_function, least_squares,
least_squares_derivative, learning_rate):
self.topology = topology
self.activate_function = activate_function
self.derivative_activate_function = derivative_activate_function
self.least_squares = least_squares
self.least_squares_derivative = least_squares_derivative
self.learning_rate = learning_rate
self.network = self.create_network()
def create_network(self):
network = []
for la, layer in enumerate(self.topology[:-1]):
network.append(NeuralLayer(self.topology[la], self.topology[la + 1], self.activate_function,
self.derivative_activate_function))
return network
def network(self):
return self.network
def train(self, x, y, train=True):
out = [(None, x)]
# Forward pass
for la, layer in enumerate(self.network):
# Weighted sum
z = out[-1][1] @ self.network[la].W + self.network[la].b
# Action activate function in current perceptron
a = self.network[la].activate_function(z)
out.append((z, a))
# Back-propagation
if train:
deltas = []
# Backward pass
for layer in reversed(range(0, len(self.network))):
# Reverse weighted sum and computed activation
a = out[layer + 1][1]
# Last layer
if layer == len(self.network) - 1:
deltas.insert(0, self.least_squares_derivative(a, y) * self.derivative_activate_function(a))
# Other layers that are not the last
else:
deltas.insert(0, deltas[0] @ _W.T * self.derivative_activate_function(a))
_W = self.network[layer].W
# Gradiant descent for bias
self.network[layer].b = self.network[layer].b - np.mean(deltas[0], axis=0,
keepdims=True) * self.learning_rate
# Gradiant descent for weights
self.network[layer].W = self.network[layer].W - out[layer][1].T @ deltas[0] * self.learning_rate
return out[-1][1]