LeNet/LeNetNetwork.cs at master · patrickmeiring/LeNet · GitHub

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
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;

namespace AdvancedOCR
{
    class LeNetNetwork
    {
        public InputStep inputLayer { get; protected set; }

        const int FirstConvolutionCount = 6;
        const int FirstConvolutionSize = 5;
        public ConvolutionStep[] FirstConvolutions { get; protected set; }
        public SubsamplingStep[] FirstSubsampling { get; protected set; }

        const int SecondConvolutionCount = 16;
        const int SecondConvolutionSize = 5;

        readonly bool[,] SecondConvolutionConnections = new bool[16,6] {
                {true ,true ,true ,false,false,false},
                {false,true ,true ,true ,false,false},
                {false,false,true ,true ,true ,false},
                {false,false,false,true ,true ,true },
                {true ,false,false,false,true ,true },
                {true ,true ,false,false,false,true },

                {true ,true ,true ,true ,false,false},
                {false,true ,true ,true ,true ,false},
                {false,false,true ,true ,true ,true },
                {true ,false,false,true ,true ,true },
                {true ,true ,false,false,true ,true },
                {true ,true ,true ,false,false,true },

                {true ,true ,false,true ,true ,false},
                {false,true ,true ,false,true ,true },
                {true ,false,true ,true ,false,true },

                {true ,true ,true ,true ,true ,true }
            };

        public ConvolutionStep[] SecondConvolutions { get; protected set; }
        public SubsamplingStep[] SecondSubsampling { get; protected set; }

        const int ConsolidationNeurons = 120;
        public const int OutputFeedForwardNeurons = 84;


        public FeedForwardStep Consolidation { get; protected set; }
        public FeedForwardStep Output { get; protected set; }
        public MarkingStep Marking { get; protected set; }

        protected Step[] forwardSteps;
        protected Step[] reverseSteps;

        protected LeNetConfiguration configuration;

        public LeNetNetwork(params char[] characters)
        {
            Weights.GlobalLearningRate = 0.00005;
            Weights.GlobalMu = 0.02;
            configuration = LeNetConfiguration.FromCharacters(characters);
            CreateNetwork();
        }

        private void CreateNetwork()
        {
            InstanciateSteps();
            CreateStepLists();
        }

        private void InstanciateSteps()
        {
            inputLayer = new InputStep(32, 32);
            FirstConvolutions = new ConvolutionStep[FirstConvolutionCount];
            FirstSubsampling = new SubsamplingStep[FirstConvolutionCount];
            for (int i = 0; i < FirstConvolutionCount; i++)
            {
                ConvolutionStep convolutionStep = new ConvolutionStep(inputLayer, FirstConvolutionSize);
                FirstConvolutions[i] = convolutionStep;
                FirstSubsampling[i] = new SubsamplingStep(convolutionStep, 2);
            }

            SecondConvolutions = new ConvolutionStep[SecondConvolutionCount];
            SecondSubsampling = new SubsamplingStep[SecondConvolutionCount];
            for (int i = 0; i < SecondConvolutionCount; i++)
            {
                RectangularStep[] inputs = FirstSubsampling.Where((item, upstreamIndex) => SecondConvolutionConnections[i, upstreamIndex]).ToArray();
                ConvolutionStep convolutionStep = new ConvolutionStep(inputs, SecondConvolutionSize);
                SecondConvolutions[i] = convolutionStep;
                SecondSubsampling[i] = new SubsamplingStep(convolutionStep, 2);
            }

            Consolidation = new FeedForwardStep(120, SecondSubsampling);

            Output = new FeedForwardStep(OutputFeedForwardNeurons, Consolidation);
            Marking = new MarkingStep(Output, configuration);
        }

        private void CreateStepLists()
        {
            List<Step> steps = new List<Step>();

            steps.AddRange(FirstConvolutions);
            steps.AddRange(FirstSubsampling);
            steps.AddRange(SecondConvolutions);
            steps.AddRange(SecondSubsampling);
            steps.Add(Consolidation);
            steps.Add(Output);
            steps.Add(Marking);

            this.forwardSteps = steps.ToArray();
            steps.Reverse();
            this.reverseSteps = steps.ToArray();
        }

        private bool preTraining;
        public bool IsPreTraining
        {
            get { return preTraining; }
            set
            {
                if (preTraining != value)
                {
                    preTraining = value;
                    Array.ForEach(forwardSteps, step => step.IsPreTraining = preTraining);
                }
            }
        }

        public void PropogateForward(DataSetItem inputs)
        {
            inputLayer.SetInputs(inputs.Inputs);
            Array.ForEach(forwardSteps, step => step.PropogateForward());
        }

        static int iterations = 0;
        public TrainingResults Train(DataSetItem inputs)
        {
            PropogateForward(inputs);
            Marking.CorrectClass = Array.IndexOf(configuration.Characters, inputs.Character);
            Array.ForEach(reverseSteps, step => step.PropogateBackwards());

            int correctOutputIndex = Array.IndexOf(configuration.Characters, inputs.Character);

            return new TrainingResults(Marking.Output, correctOutputIndex);

        }

    }

    class TrainingResults
    {
        public TrainingResults(double[] results, int correctClass)
        {
            Error = results[correctClass];
            double minimumNonCorrectError = results.Where((result, index) => index != correctClass).Min();
            Correct = minimumNonCorrectError > Error;
        }
        public readonly double Error;
        public readonly bool Correct;
    }

}