-
Notifications
You must be signed in to change notification settings - Fork 0
/
debug.cpp
269 lines (227 loc) · 9.53 KB
/
debug.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
267
268
269
#include <iostream>
#include "utils.hpp"
// debugging the iris neural net
class Neuron {
private:
double m_value; // the neuron's value
double m_activation; // the neuron's activation
double m_derived; // the derivative of the activation function
std::vector<double> m_weights; // the weights of the connections from this neuron
double m_gradient; // gradient for backpropagation
public:
// Default constructor
Neuron() : m_value(0.0), m_activation(0.0), m_derived(0.0), m_gradient(0.0) {}
// Constructor
Neuron(int numWeights) : m_value(0.0), m_activation(0.0), m_derived(0.0), m_weights(numWeights), m_gradient(0.0) {
// Initialize weights with Xavier initialization
std::random_device rd;
std::mt19937 gen(rd());
std::uniform_real_distribution<double> dist(-sqrt(6.0 / (numWeights + 1)), sqrt(6.0 / (numWeights + 1)));
for (double& weight : m_weights) {
weight = dist(gen);
}
}
// constructor
Neuron(const Neuron& other) : m_value(other.m_value), m_activation(other.m_activation), m_derived(other.m_derived), m_gradient(other.m_gradient), m_weights(other.m_weights) {}
// Getters and setters...
void setValue(double value) { m_value = value; }
double getValue() const { return m_value; }
void setActivation(double activation) { m_activation = activation; }
double getActivation() const { return m_activation; }
void setDerived(double derived) { m_derived = derived; }
double getDerived() const { return m_derived; }
void setWeight(int index, double weight) {
if (index >= 0 && index < m_weights.size()) {
m_weights[index] = weight;
}
}
double getWeight(int index) const {
if (index >= 0 && index < m_weights.size()) {
return m_weights[index];
}
return 0.0; // or some other default value
}
void setGradient(double gradient) { m_gradient = gradient; }
double getGradient() const { return m_gradient; }
};
class Layer {
private:
std::vector<Neuron> m_neurons; // neurons in this layer
public:
// Constructor
Layer(int size, int nextSize) : m_neurons(size, Neuron(nextSize)) {}
// Getters and setters for the neurons
Neuron& getNeuron(int index) {
if (index >= 0 && index < m_neurons.size()) {
return m_neurons[index];
}
return m_neurons[0]; // or some other default value
}
void setNeuron(int index, const Neuron& neuron) {
if (index >= 0 && index < m_neurons.size()) {
m_neurons[index] = neuron;
}
}
int size() const {
return m_neurons.size();
}
};
class NeuralNetwork {
private:
std::vector<Layer> m_layers; // layers in the network
public:
// Constructor
NeuralNetwork(int inputSize, int hiddenSize1, int hiddenSize2, int outputSize) {
m_layers.push_back(Layer(inputSize, hiddenSize1));
m_layers.push_back(Layer(hiddenSize1, hiddenSize2)); // 1st hidden layer
m_layers.push_back(Layer(hiddenSize2, outputSize)); // 2nd hidden layer
m_layers.push_back(Layer(outputSize, 1));
}
// Training functions:
void feedForward(const std::vector<double>& input) {
// Set the input layer
for (int i = 0; i < input.size(); ++i)
m_layers[0].getNeuron(i).setValue(input[i]);
// Forward propagation till the last hidden layer
for (int i = 1; i < m_layers.size() - 1; ++i) { // removed the == operator
for (int j = 0; j < m_layers[i].size(); ++j) {
double activation = 0.0;
for (int k = 0; k < m_layers[i - 1].size(); ++k)
activation += m_layers[i - 1].getNeuron(k).getValue() * m_layers[i - 1].getNeuron(k).getWeight(j);
m_layers[i].getNeuron(j).setValue(sigmoid(activation));
}
}
// For the output layer, apply sigmoid activation for binary classification
int outputLayerIndex = m_layers.size() - 1;
for (int j = 0; j < m_layers[outputLayerIndex].size(); ++j) {
double activation = 0.0;
for (int k = 0; k < m_layers[outputLayerIndex - 1].size(); ++k)
activation += m_layers[outputLayerIndex - 1].getNeuron(k).getValue() * m_layers[outputLayerIndex - 1].getNeuron(k).getWeight(j);
double sigmoidActivation = sigmoid(activation);
m_layers[outputLayerIndex].getNeuron(j).setValue(sigmoidActivation);
}
}
void backPropagate(const std::vector<double>& target) {
// Calculate gradient for output layer
int outputLayerIndex = m_layers.size() - 1;
for (int i = 0; i < m_layers[outputLayerIndex].size(); ++i) {
double output = m_layers[outputLayerIndex].getNeuron(i).getValue();
double targetVal = target[i];
double gradient = output - targetVal;
m_layers[outputLayerIndex].getNeuron(i).setGradient(gradient);
}
// Calculate gradient for hidden layers
for (int i = outputLayerIndex - 1; i >= 0; --i) {
for (int j = 0; j < m_layers[i].size(); ++j) {
double error = 0.0;
for (int k = 0; k < m_layers[i + 1].size(); ++k)
if (i != m_layers.size() - 2)
error += m_layers[i + 1].getNeuron(k).getGradient() * m_layers[i].getNeuron(j).getWeight(k);
double gradient = error * sigmoidDerivative(m_layers[i].getNeuron(j).getValue());
m_layers[i].getNeuron(j).setGradient(gradient);
}
}
}
double calculateError(const std::vector<double>& target) {
double totalError = 0;
int outputLayerIndex = m_layers.size() - 1;
for (int i = 0; i < m_layers[outputLayerIndex].size(); ++i) {
double output = m_layers[outputLayerIndex].getNeuron(i).getValue();
double targetVal = target[i];
totalError += targetVal * log(output + 1e-10); // Apply the softmax and cross-entropy in the error calculation
}
return -totalError;
}
void updateWeights(double learningRate) {
for (int i = 1; i < m_layers.size(); ++i) {
for (int j = 0; j < m_layers[i].size(); ++j) {
double deltaSum = 0.0; // Accumulate weight changes for this neuron
for (int k = 0; k < m_layers[i-1].size(); ++k) {
double weightChange = learningRate * m_layers[i].getNeuron(j).getGradient() * m_layers[i-1].getNeuron(k).getValue();
deltaSum += weightChange;
}
for (int k = 0; k < m_layers[i-1].size(); ++k) {
double newWeight = m_layers[i].getNeuron(j).getWeight(k) - deltaSum;
m_layers[i].getNeuron(j).setWeight(k, newWeight);
}
}
}
}
void printNetworkState() { //debug
for (int i = 0; i < m_layers.size(); ++i) {
std::cout << "Layer " << i << ":\n";
for (int j = 0; j < m_layers[i].size(); ++j) {
Neuron& neuron = m_layers[i].getNeuron(j);
std::cout << " Neuron " << j << ": Value = " << neuron.getValue() << ", Activation = " << neuron.getActivation() << ", Gradient = " << neuron.getGradient() << "\n";
if (i > 0) {
std::cout << " Weights: ";
for (int k = 0; k < m_layers[i - 1].size(); ++k) {
std::cout << neuron.getWeight(k) << " ";
}
std::cout << "\n";
}
}
std::cout << "\n";
}
std::cout << "-------------------\n";
}
void train(const std::vector<std::vector<double>>& inputs, const std::vector<std::vector<double>>& targets, int iterations, double learningRate) {
for (int i = 0; i < iterations; i++) {
double totalError = 0.0;
for (int j = 0; j < inputs.size(); j++) {
feedForward(inputs[j]);
backPropagate(targets[j]);
updateWeights(learningRate);
totalError += calculateError(targets[j]);
// Debugging statements
std::cout << "Sample: " << j << ", Total Error: " << totalError << std::endl;
printNetworkState(); // Print activations, gradients, weights, etc.
}
std::cout << "Iteration: " << i + 1 << " / " << iterations << ", Average Error: " << totalError / inputs.size() << std::endl;
}
}
std::vector<double> getOutput() {
std::vector<double> output;
int outputLayerIndex = m_layers.size() - 1;
for (int i = 0; i < m_layers[outputLayerIndex].size(); ++i) {
output.push_back(m_layers[outputLayerIndex].getNeuron(i).getValue());
}
return output;
}
};
std::vector<double> normalize(const std::vector<double>& input) {
double minVal = *std::min_element(input.begin(), input.end());
double maxVal = *std::max_element(input.begin(), input.end());
std::vector<double> normalizedInput;
for (double val : input) {
normalizedInput.push_back((val - minVal) / (maxVal - minVal));
}
return normalizedInput;
}
int main() {
// Initialize the net
NeuralNetwork nn(3, 2, 4, 1);
int iterations = 1000; // modify if necessary
double learningRate = 0.01;
std::vector<std::vector<double>> inputs = {
{5.1, 3.5, 1.4, 0.2}, // Iris setosa
{7.0, 3.2, 4.7, 1.4}, // Iris versicolor
{6.3, 3.3, 6.0, 2.5} // Iris virginica
};
for (std::vector<double>& input : inputs) {
input = normalize(input);
}
std::vector<std::vector<double>> targets = {
{1.0, 0.0, 0.0}, // Iris setosa
{0.0, 1.0, 0.0}, // Iris versicolor
{0.0, 0.0, 1.0} // Iris virginica
};
nn.train(inputs, targets, iterations, learningRate);
std::vector<double> output = nn.getOutput();
std::cout << "Output: ";
for (double val : output) {
std::cout << val << " ";
}
std::cout << std::endl;
return 0;
}