-
Notifications
You must be signed in to change notification settings - Fork 0
/
NeuralNetwork.py
240 lines (205 loc) · 6.33 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
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
import numpy as np
import json
import Utils
from Utils import *
# see https://towardsdatascience.com/lets-code-a-neural-network-in-plain-numpy-ae7e74410795
# see https://blog.zhaytam.com/2018/08/15/implement-neural-network-backpropagation/
def sigmoid(x):
return 1 / (1 + np.exp(-x))
def sigmoid_derivative(x):
return x * (1 - x)
def tanh(x):
return np.tanh(x)
def tanh_derivative(x):
return 1 - x ** 2
def relu_derivative(x):
x[x <= 0] = 0
x[x > 0] = 1
return x
def relu(Z):
return np.maximum(0,Z)
def relu_backward(dA, Z):
dZ = np.array(dA, copy = True)
dZ[Z <= 0] = 0
return dZ
def make_bias(n):
# step = 0.1
# van = step
# tot = n * step
# t = np.arange(van, tot + step / 10, step)
# return np.ones(n)
# return t
return np.random.randn(n) * 0.1
def make_weights(n):
# step = 0.1
# van = step
# tot = n[0] * n[1] * step
# t = np.arange(van, tot + step / 10, step)
# t = t.reshape(n)
# return np.ones(n)
# return t
return np.random.randn(n[0], n[1]) * 0.1
class Layer:
nodes = 0
input_nodes = None
bias = None
weights = None
last_activation = None
# activation_funcname = "sigmoid"
activation_funcname = "relu"
def __init__(self, inn, n, af):
"""
:param int inn: the number of input nodes
:param int n: the number of nodes in this layer
"""
self.input_nodes = inn
self.nodes = n
self.weights = make_weights((self.input_nodes, self.nodes))
self.bias = make_bias((self.nodes))
self.activation_funcname = af
def activate(self, x):
r = np.dot(x, self.weights) + self.bias
if self.activation_funcname == "sigmoid":
r = sigmoid(r)
elif self.activation_funcname == "relu":
r = relu(r)
elif self.activation_funcname == "tanh":
r = tanh(r)
else:
print("Unknown activation function: {}".format(self.activation_funcname))
self.last_activation = r
# print(r)
return self.last_activation
def show(self):
print("Layer: ", self.input_nodes, " -> ", self.nodes)
print("Weights")
print(self.weights)
print("Bias")
print(self.bias)
print("Activation Function")
print(self.activation_funcname)
def mutate(self, rate):
ff = np.vectorize(lambda x : ( x + np.random.normal(0, 0.1) if np.random.random() < rate else x))
self.weights = ff(self.weights)
self.bias = ff(self.bias)
def serialize(self):
return {
"input_nodes": self.input_nodes,
"nodes": self.nodes,
"activation_function": self.activation_funcname,
"bias": self.bias.tolist(),
"weights": self.weights.tolist()
}
@staticmethod
def deserialize(data):
layer = Layer(data["input_nodes"], data["nodes"], data["activation_function"])
layer.last_activation = None
layer.bias = np.array(data["bias"])
layer.weights = np.array(data["weights"])
# layer.show()
return layer
class NeuralNetwork:
input_nodes = 1
output_nodes = 1
hidden_layers = 1
hidden_nodes = 1
layers = []
def __init__(self, inn, hl, hn, oun):
"""
:param int inn: the number of input nodes
:param int hl: the number of hidden layers
:param int hn: the number of nodes in each hidden layer
:param int oun: the number of output nodes
"""
self.layers = []
self.input_nodes = inn
self.output_nodes = oun
self.hidden_layers = hl
self.hidden_nodes = hn
self.add_layer(self.input_nodes, self.hidden_nodes, "sigmoid")
for i in range(0, self.hidden_layers - 1):
self.add_layer(self.hidden_nodes, self.hidden_nodes, "sigmoid")
self.add_layer(self.hidden_nodes, self.output_nodes, "relu")
#for i in range(0, len(self.layers)):
# self.layers[i].show()
def add_layer(self, inn, n, af):
layer = Layer(inn, n, af)
self.layers.append(layer)
def feed_forward(self, x):
for layer in self.layers:
x = layer.activate(x)
return x
def predict(self, x):
y = self.feed_forward(x)
return y
def backpropagation(self, x, y, learning_rate):
output = self.feed_forward(x)
for i in reversed(range(len(self.layers))):
layer = self.layers[i]
if layer == self.layers[-1]:
layer.error = y - output
layer.delta = layer.error * sigmoid_derivative(output)
else:
next_layer = self.layers[i + 1]
layer.error = np.dot(next_layer.weights, next_layer.delta)
layer.delta = layer.error * sigmoid_derivative(layer.last_activation)
for i in range(len(self.layers)):
layer = self.layers[i]
input_to_use = np.atleast_2d(x if i == 0 else self.layers[i - 1].last_activation)
layer.weights += layer.delta * input_to_use.T * learning_rate
def train(self, X, y, learning_rate, max_epochs):
"""
Trains the neural network using backpropagation.
:param X: The input values.
:param y: The target values.
:param float learning_rate: The learning rate (between 0 and 1).
:param int max_epochs: The maximum number of epochs (cycles).
:return: The list of calculated MSE errors.
"""
mses = []
for i in range(max_epochs):
for j in range(len(X)):
self.backpropagation(X[j], y[j], learning_rate)
if i % 10 == 0:
mse = np.mean(np.square(y - self.feed_forward(X)))
mses.append(mse)
print('Epoch: #%s, MSE: %f' % (i, float(mse)))
return mses
def copy(self):
new = NeuralNetwork(self.input_nodes, self.hidden_layers, self.hidden_nodes, self.output_nodes)
for i in range(len(self.layers)):
new.layers[i].weights = self.layers[i].weights.copy()
new.layers[i].bias = self.layers[i].bias.copy()
new.layers[i].activation_funcname = self.layers[i].activation_funcname
return new
def mutate(self, rate):
for i in range(len(self.layers)):
self.layers[i].mutate(rate);
def serialize(self):
data = {}
data["input_nodes"] = self.input_nodes
data["output_nodes"] = self.output_nodes
data["hidden_layers"] = self.hidden_layers
data["hidden_nodes"] = self.hidden_nodes
data["layers"] = []
for i in range(len(self.layers)):
data["layers"].append(self.layers[i].serialize())
return json.dumps(data, indent=4)
@staticmethod
def deserialize(jsonstring):
data = json.loads(jsonstring)
nn = NeuralNetwork(data["input_nodes"], data["hidden_layers"], data["hidden_nodes"], data["output_nodes"])
nn.layers.clear()
for p in data['layers']:
layer = Layer.deserialize(p)
nn.layers.append(layer)
return nn
#@staticmethod
#def accuracy(y_pred, y_true):
# """
# Calculates the accuracy between the predicted labels and true labels.
# :param y_pred: The predicted labels.
# :param y_true: The true labels.
# :return: The calculated accuracy.
# """
# return (y_pred - y_true).mean()