-
Notifications
You must be signed in to change notification settings - Fork 0
/
Example_LEGO.cpp
116 lines (89 loc) · 3.25 KB
/
Example_LEGO.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
//
// Created by pbo on 18.06.19.
//
#include <iostream>
#include <Placeholder.hpp>
#include <Operation.hpp>
#include <SummationOp.hpp>
#include <MultiplicationOp.hpp>
#include <DataInitialization.hpp>
#include <Parameter.hpp>
#include <ConvolutionOp.hpp>
#include <ReLuOp.hpp>
#include <MaxPoolOp.hpp>
#include <SoftmaxOp.hpp>
#include <CrossEntropyOp.hpp>
#include "mnist/mnist_reader.hpp"
#include <mnist/mnist_utils.hpp>
#include <lodepng.hpp>
#include <iomanip>
#include <IO.hpp>
#include <InputLayer.hpp>
#include <ConvolutionLayer.hpp>
#include <AbstractLayer.hpp>
#include <MaxPoolLayer.hpp>
#include <DenseLayer.hpp>
#include <LossLayer.hpp>
#include <LogitsLayer.hpp>
#include <NeuralNetwork.hpp>
#include <filesystem>
#include <algorithm>
#include <random>
#include <LegoDataLoader.hpp>
int main() {
//
/*
* batch_size: if this is changed '#define BATCH_SIZE' in Node.hpp has to be changed as well
* epochs: sets the amount of epochs for training
* amount_batches: 'batch_size*amount_batches' gives the total amount of samples
* writeWeights: if set the trained Weights are written to Source_Directory/WeightDeposit
* readWeights: if set (and Weights have already been Written once) weights are initialized with weights from Source_Directory/WeightDeposit
*/
int batch_size = 16;
int epochs = 5;
double learningRate = 0.0001;
int amount_batches = 10;
bool writeWeights = true;
bool readWeights = false;
std::vector<std::pair<std::string, int>> shuffledDataStrings = LegoDataLoader::shuffleData(DATA_DIR);
DataSet legoData;
LegoDataLoader::getData(batch_size * amount_batches, shuffledDataStrings, legoData);
/*
* Create Neural Network
*/
HyperParameters config(epochs, batch_size, learningRate);
std::shared_ptr<Graph> graph = std::make_shared<Graph>();
//Create InputLayer
auto inputLayer = std::make_shared<InputLayer>(graph, batch_size, 160000, 4);
//Convolutional Layer 1
auto convolution1 = std::make_shared<ConvolutionLayer>(inputLayer, graph, ActivationType::ReLu,
32, 8, 2, InitializationType::Xavier);
auto maxPool1 = std::make_shared<MaxPoolLayer>(convolution1, graph, 2, 2);
//convolutional Layer 2
auto convolution2 = std::make_shared<ConvolutionLayer>(maxPool1, graph, ActivationType::ReLu, 64, 5, 2, InitializationType::Xavier);
//Maxpooling
auto maxPool2 = std::make_shared<MaxPoolLayer>(convolution2, graph, 2, 2);
//Dense Layer 1
auto dense1 = std::make_shared<DenseLayer>(maxPool2, graph, ActivationType::ReLu, 1024, InitializationType::Xavier);
//Dense Layer 2
auto dense2 = std::make_shared<DenseLayer>(dense1, graph, ActivationType::None, 16, InitializationType::Xavier);
//Logits Layer
auto logits = std::make_shared<LogitsLayer>(dense2, graph, 16);
// Cost Layer
auto loss = std::make_shared<LossLayer>(logits, graph, LossType::CrossEntropy);
//Create Deep Learning session
NeuralNetwork network(graph, inputLayer, loss);
/*
* Initialize Network with precalculated Weights
*/
if (readWeights) { network.readParameters(STORAGE, "lego_layer"); }
/*
* Train the Network
*/
TrainingEvaluation eval = network.trainAndValidate(legoData, config);
// network.train(legoData,config);
/*
* Write calculated Weights to Network
*/
if (writeWeights) { network.writeParameters(STORAGE, "lego_layer"); }
}