-
Notifications
You must be signed in to change notification settings - Fork 5
/
Solve_Xor.cpp
266 lines (224 loc) · 5.2 KB
/
Solve_Xor.cpp
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
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
//Hunar @ Brainxyz.com
#pragma once
#include <iostream>
#include <vector>
#include <math.h>
//////// helper functions;
double gen_weight() {
return ((double)rand() / (double)RAND_MAX) - 0.5;
}
double activationF(double val) {
return std::tanh(val);
}
//// class declarations
struct Edge;
struct Node;
struct Layer;
struct Network;
////// class function declarations
struct Edge {
double fW = gen_weight();
double cW = fW;
Node* from = nullptr;
Node* to = nullptr;
void mutateEdge(double lr);
void childBecomesFather();
};
struct Node {
double fY = 1;
double cY = 1;
std::vector<Edge*> from_edges;
std::vector<Edge*> to_edges;
void spreadOut();
void mutateChildern(double lr);
void connectTo(Network* net, Node* targNode, double weight);
};
struct Layer {
std::vector<Node*> nodes;
void appendNodes(long num);
void resetLayer();
void spreadOut(double lr, Network* net);
void activate();
void fconnectTo(Network* net, Layer* targLayer);
void sconnectTo(Network* net, Layer* targLayer, long sparse_rate);
};
struct Network {
std::vector<Edge*> edgeList;
std::vector<Edge*> activeEdges;
Layer sensor = Layer();
Layer hidden = Layer();
Layer out = Layer();
Layer bias1 = Layer();
Network(long, long, long);
void fullyConnect();
void sparseConnect(std::vector<double>);
void setInput(std::vector<double> inputs);
void forward(std::vector<double> inputs, double lr);
void updateWeights();
void mutateWeights(double lr);
void resetNet();
};
//// implementations //////
void Edge::mutateEdge(double lr) {
cW = fW + (gen_weight() * lr);
}
void Edge::childBecomesFather() { fW = cW; }
void Node::spreadOut() {
for (auto e : to_edges) {
e->to->fY += fY * e->fW;
e->to->cY += cY * e->cW;
}
}
void Node::mutateChildern(double lr) {
for (auto e : to_edges) {
e->mutateEdge(lr);
}
}
void Node::connectTo(Network* net, Node* targNode, double weight) {
auto edge = new Edge();
edge->fW = weight;
edge->cW = weight;
edge->from = this;
edge->to = targNode;
net->edgeList.push_back(edge);
to_edges.push_back(edge);
targNode->from_edges.push_back(edge);
}
void Layer::appendNodes(long num) {
for (long i = 0; i < num; i++) {
nodes.push_back(new Node());
}
}
void Layer::resetLayer() {
for (auto n : nodes) {
n->fY = 0;
n->cY = 0;
}
}
void Layer::spreadOut(double lr, Network* net) {
for (auto n : nodes) {
if (n->fY > 0) {
n->spreadOut();
for (auto e : n->to_edges) {
net->activeEdges.push_back(e);
}
}
}
}
void Layer::activate()
{
for (auto n : nodes) {
n->fY = activationF(n->fY);
n->cY = activationF(n->cY);
}
}
void Layer::fconnectTo(Network* net, Layer* targLayer)
{
for (auto sn : nodes) {
for (auto tn : targLayer->nodes) {
sn->connectTo(net, tn, gen_weight());
}
}
}
void Layer::sconnectTo(Network* net, Layer* targLayer, long sparse_rate)
{
for (auto sn : nodes) {
for (auto tn : targLayer->nodes) {
if ((gen_weight() + 0.5) < sparse_rate) {
sn->connectTo(net, tn, gen_weight());
}
}
}
}
Network::Network(long L1, long L2, long L3) {
sensor.appendNodes(L1);
hidden.appendNodes(L2);
out.appendNodes(L3);
bias1.appendNodes(1);
}
void Network::fullyConnect()
{
sensor.fconnectTo(this, &hidden);
hidden.fconnectTo(this, &out);
bias1.fconnectTo(this, &out);
}
void Network::sparseConnect(std::vector<double> amount)
{
sensor.sconnectTo(this, &hidden, amount[0]);
hidden.sconnectTo(this, &out, amount[1]);
bias1.fconnectTo(this, &out);
}
void Network::setInput(std::vector<double> inputs)
{
for (long i = 0; i < inputs.size(); i++) {
sensor.nodes[i]->fY = inputs[i];
sensor.nodes[i]->cY = inputs[i];
}
}
void Network::mutateWeights(double lr) {
for (auto e : edgeList) {
e->mutateEdge(lr);
}
}
void Network::forward(std::vector<double> inputs, double lr)
{
setInput(inputs);
sensor.spreadOut(lr, this);
hidden.activate();
hidden.spreadOut(lr, this);
bias1.spreadOut(lr, this);
}
void Network::updateWeights()
{
for (auto e : activeEdges) {
e->childBecomesFather();
}
}
void Network::resetNet()
{
sensor.resetLayer();
hidden.resetLayer();
out.resetLayer();
activeEdges.clear();
}
int main() {
long epochs = 1000;
const float lr = 0.1;
const int ins = 2;
const int nodes = 10;
const int out = 1;
const int nsamples = 4;
Network net = Network(ins, nodes, out);
net.fullyConnect();
//net.sparseConnect({ 0.1, 0.5 });
// data & labels
std::vector<std::vector<double>> inputs = { {0, 0} ,{1, 1},{1, 0},{0, 1} };
std::vector<double> labels= { 0, 0, 1, 1 };
// training loop
double fE, cE, label;
std::vector<double> inp;
for (int i = 0; i < epochs; i++) {
fE = 0; cE = 0; // initialize father Erorr(fE) and child Error(cE)
for (int j = 0; j < nsamples; j++) {
inp = inputs[j];
label = labels[j];
net.resetNet();
net.forward(inp, lr);
// calculates the network error
fE = fE + std::abs(net.out.nodes[0]->fY - label);
cE = cE + std::abs(net.out.nodes[0]->cY - label);
}
if (fE > cE)
net.updateWeights();
net.mutateWeights(lr);
}
std::cout << "assess training" << std::endl;
for (int j = 0; j < labels.size(); j++) {
net.resetNet();
inp = inputs[j];
label = labels[j];
net.forward(inp, lr);
std::cout << "target:" << label << " pred:" << net.out.nodes[0]->fY << std::endl;
}
std::cin.get();
}