private void Awake()
{
instance = this;
thisNet = new NeuralNetwork.NeuralNet(4, 3, 1);
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 1.0, 1.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 0.0, 1.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 1.0, 0.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 0.0, 0.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 1.0, 1.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 0.0, 1.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 1.0, 0.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 0.0, 0.0, 0.5 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 1.0, 1.0, 0.1 }, new double[] { 0.1 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 1.0, 0.0, 0.1 }, new double[] { 0.1 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 0.5, 0.0, 0.1 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 0.5, 1.0, 0.1 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 1.0, 1.0, 0.1 }, new double[] { 0.1 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 1.0, 0.0, 0.1 }, new double[] { 0.1 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 0.5, 0.0, 0.1 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 0.5, 1.0, 0.1 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 1.0, 1.0, 0.9 }, new double[] { 0.9 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 0.0, 1.0, 0.9 }, new double[] { 0.9 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 1.0, 0.5, 0.9 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 1.0, 0.0, 0.5, 0.9 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 1.0, 1.0, 0.9 }, new double[] { 0.9 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 0.0, 1.0, 0.9 }, new double[] { 0.9 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 1.0, 0.5, 0.9 }, new double[] { 0.5 }));
DataSets.Add(new NeuralNetwork.DataSet(new double[] { 0.0, 0.0, 0.5, 0.9 }, new double[] { 0.5 }));
thisNet.Train(DataSets, 10);
}
public NeuralNet Crossover(NeuralNet partner)
{
OutputLayer.Crossover(partner.OutputLayer);
foreach (var layer in HiddenLayers.Zip(partner.HiddenLayers))
{
layer.Key.Crossover(layer.Value);
}
return(this);
}
static void Main(string[] args)
{
var TrainingData = MNIST.ReadMNIST(false);
//Set Random Seed for better debugging
Accord.Math.Random.Generator.Seed = 5;
//Input, Hidden, Output
NeuralNet NN = new NeuralNet(2, 2, 1);
TrainXORTest(NN);
}
private void OnEnable()
{
net = new NeuralNet(3, 1, 1);
dataSets = new List <DataSet>();
Debug.Log("inicializando");
double[] inputs = new double[3];
double[] desireOutputs = new double[1];
inputs[0] = 0.1f;
inputs[1] = 0.1f;
inputs[2] = 0.1f;
desireOutputs[0] = 0.1f;
dataSets.Add(new DataSet(inputs, desireOutputs));
net.Train(dataSets, 0.001f);
double[] vals = new double[3];
vals[0] = 0.1f;
vals[1] = 0.1f;
vals[2] = 0.9f;
double[] respuesta = net.Compute(vals);
Debug.Log(respuesta[0]);
if (respuesta[0] < 0.09f)
{
col = 0;
}
if (respuesta[0] > 0.09f && respuesta[0] < 0.1)
{
col = 1;
}
if (respuesta[0] > 0.1f)
{
col = 2;
}
sonidoNotificacion.SetActive(true);
indice++;
botones[0].SetActive(true);
botones[1].SetActive(true);
botones[2].SetActive(true);
cajasMensaje[indice - 1].SetActive(true);
textoMensaje[indice - 1].text = maquina.mensajesMaquina[indice - 1, col];
textoRespuesta[0].text = maquina.mensajesUsuario[indice - 1, 0];
textoRespuesta[1].text = maquina.mensajesUsuario[indice - 1, 1];
textoRespuesta[2].text = maquina.mensajesUsuario[indice - 1, 2];
if (indice == 6)
{
maquina.ActivarEstado(maquina.EstadoCaptura);
}
}
//Crossover
public void Crossover(NeuralNet other, Random rng)
{
for (int i = 0; i < Layers.Count; i++)
{
for (int j = 0; j < Layers[i].Neurons.Length; j++)
{
if (rng.Next(2) == 0)
{
Layers[i].Neurons[j].BiasWeight = other.Layers[i].Neurons[j].BiasWeight;
other.Layers[i].Neurons[j].Weights.CopyTo(Layers[i].Neurons[j].Weights, 0);
}
}
}
}
//Initialize Network and Dataset
void Awake()
{
int numInputs, numHiddenLayers, numOutputs;
if (File.Exists("data.txt"))
{
net = ReadFromXmlFile <NeuralNet>("data.txt");
Debug.Log("THIS WORKS");
}
else
{
net = new NeuralNet(37, 2, 10);
Debug.Log("reached here");
}
dataSets = new List <DataSet>();
}
static void Main(string[] args)
{
NeuralNet net = new NeuralNet(new int[] { 2, 3, 1 });
double[][][] trainData = net.GetTrainData("trainSet.txt");
net.OpenNet("1.xml");
//net.Training(trainData);
//net.SaveNet("1.xml");
Console.WriteLine("in: 0 0\nout: {0}\n\nin: 1 1\nout: {1}\n\nin: 0 1\nout: {2}\n\nin: 1 0\nout: {3}\n\n",
net.Ask(new double[] { 0, 0 })[0],
net.Ask(new double[] { 1, 1 })[0],
net.Ask(new double[] { 0, 1 })[0],
net.Ask(new double[] { 1, 0 })[0]);
Console.ReadKey();
}
static void Main(string[] args)
{
//First argument will be the path to the images
//Second argument will be the path to the labels
int width = 28;
int height = 28;
List <Digit> digits = GetLabeledImages(args[0], args[1], width, height).Shuffle().Cast <Digit>().ToList();
NeuralNet net = NeuralNet.Input(width * height).Layer <ReLU>(20).Layer <LeakyReLU1>(20).Output <Sigmoid>(10);
IEnumerable <IDataSet> training = digits.Take(digits.Count() - 100);
IEnumerable <IDataSet> test = digits.Skip(digits.Count() - 100);
net.TrainByExample(training, epochs: 2, fuzz: 0.2d);
Console.WriteLine($"\n{test.Where(sample => net.Evaluate(sample.Values).Select((o, i) => (o, i)).MaxBy(o => o.o).i == sample.Targets.MaxIndex()).Count()}/{test.Count()} correct from test set.");
File.WriteAllText(@"C:\NNTrimmed.json", net.ToJson());
}
private static void TrainXORTest(NeuralNet NN)
{
double[][] TrainData =
{
new double[] { 0, 1 },
new double[] { 1, 1 },
new double[] { 1, 0 },
new double[] { 0, 0 },
};
var Labels = new double[][]
{
new double[] { 0 },
new double[] { 1 },
new double[] { 0 },
new double[] { 0 },
};
for (int i = 1; i < 10000; i++)
{
TrainData.Shuffle();
NN.Train(TrainData, Labels, 0.01);
}
}
/// <summary>
/// Creates the visual representation of what an input is doing to the neural network
/// </summary>
/// <param name="input"></param>
/// <returns></returns>
public List <double> NodeUpdate(List <double> input, NeuralNet nNetwork)
{
List <double> outputs = new List <double>();
int weight = 0;
if (input.Count != numInputs)
{
return(outputs);
}
//Set Input Nodes
for (int i = 0; i < input.Count; i++)
{
if (input[i] > 0)
{
inputObj[i].color = Color.green * (float)(input[i]);
inputObj[i].transform.localScale = Vector3.one * ((float)input[i]);
}
else
{
inputObj[i].color = Color.red * (float)(input[i] * -1);
inputObj[i].transform.localScale = Vector3.one * ((float)input[i] * -1);
}
}
for (int i = 0; i < numHiddenLayers + 1; i++)
{
if (i > 0)
{
input = outputs;
}
outputs = new List <double>();
weight = 0;
for (int j = 0; j < nNetwork.layers[i].numNeurons; j++)
{
double netInput = 0;
int nNumInputs = nNetwork.layers[i].neurons[j].numInputs;
//For each Weight
for (int k = 0; k < nNumInputs; k++)
{
//Sum the weights and inputs
netInput += nNetwork.layers[i].neurons[j].weights[k] * input[weight];
//Set Lines
float line = (float)(nNetwork.layers[i].neurons[j].weights[k] * input[weight++]);
if (line > 0)
{
weightObj[i][j][k].SetColors(Color.green * line, Color.green * line);
weightObj[i][j][k].SetWidth(line * .25f, line * .25f);
}
else
{
weightObj[i][j][k].SetColors(Color.red * -line, Color.red * -line);
weightObj[i][j][k].SetWidth(-line * .25f, -line * .25f);
}
}
//Add bias
netInput += nNetwork.layers[i].neurons[j].weights[nNumInputs] * -1;
double sig = NeuralNet.Sigmoid(netInput, 10);
//Set Node Color
layersObj[i][j].color = (Color.green * (float)sig);
layersObj[i][j].transform.localScale = (Vector3.one * (float)sig);
outputs.Add(sig);
weight = 0;
}
}
return(outputs);
}
public TrainAndTestPage(NeuralNetRunType runtype, NeuralNet neuralnet, MNISTDataManager mnistdata)
{
// Define GUI
// Page properties
Title = string.Format("{0} Net", runtype.ToString().ToUpperFirstOnly());
BackgroundColor = Color.SteelBlue;
Padding = new Thickness(Application.Current.MainPage.Width * 0.05, Application.Current.MainPage.Height * 0.05);
// Label for description
Label description = new Label
{
Text = string.Format("Please define the number of data sets from the MNIST {0} data base to be used for {0}ing. Each data set represents one digit. The {0} data base holds about {1:N0} data sets in total.", runtype.ToString(), mnistdata.CountData),
Margin = new Thickness(0, 0, 0, Application.Current.MainPage.Height * 0.075)
};
// Label for number of training / testing sets used
Label label_useddatasets = new Label
{
Text = String.Format("{0:N0} data set{1} selected", mnistdata.UsedDataSets, mnistdata.UsedDataSets == 0 ? "" : "s"),
Margin = new Thickness(0, 0, 0, Application.Current.MainPage.Height * 0.075)
};
// Slider for number of data sets used for training / testing
Slider slider_useddatasets = new Slider(1, mnistdata.CountData, mnistdata.UsedDataSets);
slider_useddatasets.ValueChanged += (s, e) =>
{
mnistdata.UsedDataSets = (int)e.NewValue;
label_useddatasets.Text = String.Format("{0:N0} data set{1} selected", mnistdata.UsedDataSets, mnistdata.UsedDataSets == 0 ? "" : "s");
};
// Progress bar
ProgressBar progressbar = new ProgressBar {
Progress = 0
};
// Label for showing progress
Label label_progress = new Label
{
Text = "Currently no calculation running",
};
// Button to start training / testing
Button button = new Button
{
Text = string.Format("Start {0}ing", runtype.ToString().ToUpperFirstOnly()),
WidthRequest = Application.Current.MainPage.Width * 0.5,
HeightRequest = Application.Current.MainPage.Height * 0.10
};
// Define page
Content = new ScrollView
{
Content = new StackLayout
{
Children =
{
description,
label_useddatasets,
slider_useddatasets,
button,
progressbar,
label_progress
}
}
};
// Button event
// Here training and test sessions are executed
button.Clicked += async(s, e) =>
{
// Check if another calculation is currently running
if (activerun == true)
{
var answer = await DisplayAlert("Warning", "Do you want to stop current calculation?", "Yes", "No");
if (answer)
{
endrun = true;
}
return;
}
// Check if selected data set is big
if (mnistdata.UsedDataSets > mnistdata.UsedDataSetsWarning)
{
bool answer = await DisplayAlert("Warning", "Please be aware that big data sets effect run time. Continue?", "Yes", "No");
if (!answer)
{
return;
}
}
// Set activerun and endrun flags
activerun = true;
endrun = false;
// Disable Back button
NavigationPage.SetHasBackButton(this, false);
// Rename Button to Stop
button.Text = string.Format("Stop {0}ing", runtype.ToString().ToUpperFirstOnly());
// Define parameters for progress bar calculation (increase bar 100 times or each 20 runs)
int progresssteps = Math.Max(100, mnistdata.UsedDataSets / 20);
int progressspan = (int)Math.Ceiling((double)mnistdata.UsedDataSets / progresssteps);
// Counter for training / testing runs
int i = 0;
// Initialize progress bar
label_progress.Text = i.ToString() + " / " + mnistdata.UsedDataSets.ToString();
await progressbar.ProgressTo((double)i / mnistdata.UsedDataSets, 1, Easing.Linear);
// Train neural net
if (runtype == NeuralNetRunType.train)
{
while (i < mnistdata.UsedDataSets && endrun == false)
{
// Counter for runs in progress step
int j = 0;
// Training loop for one progress step done as task
await Task.Run(() =>
{
while (j < progressspan && i < mnistdata.UsedDataSets)
{
// Train net with data set i
neuralnet.Train(mnistdata.Input(i), mnistdata.Output(i));
j++;
i++;
}
// Remember number of training data sets
neuralnet.TrainingDataCounter += j;
});
// Update progress bar
label_progress.Text = i.ToString() + " / " + mnistdata.UsedDataSets.ToString();
await progressbar.ProgressTo((double)i / mnistdata.UsedDataSets, 1, Easing.Linear);
}
// Show mesage
await DisplayAlert("Result", string.Format("Neural net trained with {0:N0} data sets", i), "OK");
}
if (runtype == NeuralNetRunType.test)
{
// Setup scorecard for capturing test results
// digittotal counts all trained figures, digitcorrect the ones that are identified correctly
Vector <double> digittotal = Vector <double> .Build.Dense(10);
Vector <double> digitcorrect = Vector <double> .Build.Dense(10);
// Testing loop
while (i < mnistdata.UsedDataSets && endrun == false)
{
// Counter for runs in progress step
int j = 0;
// Testing loop for one progress step done as task
await Task.Run(() =>
{
while (j < progressspan && i < mnistdata.UsedDataSets)
{
// Increase counter for testet digit (0..9)
int number = mnistdata.Number(i);
digittotal[number]++;
// Ask net
Vector <double> answer = neuralnet.Query(mnistdata.Input(i));
// Check if it's correct
if (answer.AbsoluteMaximumIndex() == number)
{
digitcorrect[number]++;
}
j++;
i++;
}
// Remember number of training data sets
neuralnet.TrainingDataCounter += j;
});
// Update progress bar
label_progress.Text = i.ToString() + " / " + mnistdata.UsedDataSets.ToString();
await progressbar.ProgressTo((double)i / mnistdata.UsedDataSets, 1, Easing.Linear);
}
// Remeber performance result
neuralnet.Performance.Add(digitcorrect.Sum() / digittotal.Sum());
// Show test results
await Navigation.PushAsync(new ResultsMNISTPage(neuralnet, mnistdata, digittotal, digitcorrect));
}
// Reset progress bar
label_progress.Text = "Currently no calculation running";
await progressbar.ProgressTo(0.0, 250, Easing.Linear);
// Rename Button to Start
button.Text = string.Format("Start {0}ing", runtype.ToString().ToUpperFirstOnly());
// Enable Back button
NavigationPage.SetHasBackButton(this, true);
// Cancel activerun flag
activerun = false;
};
}
public void NeuralNetConstructorAssignsCorrectNumberOfWeightsToEachLayer()
{
Layer input = new Layer(LayerType.Input, null, 0, 3);
Layer hidden = new Layer(LayerType.Hidden, new LogisticSigmoidActivationFunction(), 1, 4);
Layer output = new Layer(LayerType.Output, new SoftmaxActivationFunction(), 2, 2);
input.Nodes[0].Input = 1;
input.Nodes[1].Input = 2;
input.Nodes[2].Input = 3;
input.Nodes[0].ActivatedSum = 1;
input.Nodes[1].ActivatedSum = 2;
input.Nodes[2].ActivatedSum = 3;
List<Layer> layers = new List<Layer>();
layers.Add(input);
layers.Add(hidden);
layers.Add(output);
//24 weights
List<double> weights = new List<double>()
{
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1
};
//6 biases
List<double> biases = new List<double>()
{
1,
1,
1,
1,
1,
1
};
NeuralNet nn = new NeuralNet(layers, weights, biases);
Assert.AreEqual(nn.Layers[0].Nodes.Count * nn.Layers[1].Nodes.Count, nn.Layers[0].Weights.Count);
Assert.AreEqual(nn.Layers[1].Nodes.Count * nn.Layers[2].Nodes.Count, nn.Layers[1].Weights.Count);
}
public void NeuralNetConstructorOrdersLayersAscendingByLayerOrder()
{
Layer input = new Layer(LayerType.Input, null, 0, 3);
Layer output = new Layer(LayerType.Output, new SoftmaxActivationFunction(), 2, 2);
Layer hidden = new Layer(LayerType.Hidden, new LogisticSigmoidActivationFunction(), 1, 4);
input.Nodes[0].Input = 1;
input.Nodes[1].Input = 2;
input.Nodes[2].Input = 3;
input.Nodes[0].ActivatedSum = 1;
input.Nodes[1].ActivatedSum = 2;
input.Nodes[2].ActivatedSum = 3;
List<Layer> layers = new List<Layer>();
layers.Add(input);
layers.Add(output);
layers.Add(hidden);
//24 weights
List<double> weights = new List<double>()
{
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1,
1
};
//6 biases
List<double> biases = new List<double>()
{
1,
1,
1,
1,
1,
1
};
NeuralNet nn = new NeuralNet(layers, weights, biases);
//nn.Run();
Assert.AreEqual(0, nn.Layers[0].LayerOrder);
Assert.AreEqual(1, nn.Layers[1].LayerOrder);
Assert.AreEqual(2, nn.Layers[2].LayerOrder);
}
public ResultsMNISTPage(NeuralNet neuralnet, MNISTDataManager mnistdata, Vector <double> digittotal, Vector <double> digitcorrect)
{
// Define GUI
// Page properties
Title = "Results";
BackgroundColor = Color.SteelBlue;
Padding = new Thickness(Application.Current.MainPage.Width * 0.05, Application.Current.MainPage.Height * 0.05);
// Define styles for labels in grid
var labelgridStyle = new Style(typeof(Label))
{
Setters =
{
new Setter {
Property = Label.TextColorProperty, Value = Color.GhostWhite
},
new Setter {
Property = Label.BackgroundColorProperty, Value = Color.SteelBlue
},
new Setter {
Property = Label.HorizontalOptionsProperty, Value = LayoutOptions.FillAndExpand
},
new Setter {
Property = Label.HorizontalTextAlignmentProperty, Value = TextAlignment.Center
},
new Setter {
Property = Label.VerticalOptionsProperty, Value = LayoutOptions.FillAndExpand
},
new Setter {
Property = Label.VerticalTextAlignmentProperty, Value = TextAlignment.Center
}
}
};
var labelgridhighlightStyle = new Style(typeof(Label))
{
BasedOn = labelgridStyle,
Setters =
{
new Setter {
Property = Label.TextColorProperty, Value = Color.GhostWhite
},
new Setter {
Property = Label.BackgroundColorProperty, Value = Color.DarkOrange
},
new Setter {
Property = Label.FontAttributesProperty, Value = FontAttributes.Bold
}
}
};
var labelgridheadlineStyle = new Style(typeof(Label))
{
BasedOn = labelgridStyle,
Setters =
{
new Setter {
Property = Label.TextColorProperty, Value = Color.SteelBlue
},
new Setter {
Property = Label.BackgroundColorProperty, Value = Color.GhostWhite
},
new Setter {
Property = Label.FontAttributesProperty, Value = FontAttributes.Bold
}
}
};
// Define 11x3 grid for representing results
Grid grid = new Grid();
grid.BackgroundColor = Color.GhostWhite;
grid.ColumnSpacing = 1;
grid.RowSpacing = 1;
grid.Padding = new Thickness(1);
grid.HorizontalOptions = LayoutOptions.FillAndExpand;
grid.VerticalOptions = LayoutOptions.FillAndExpand;
for (int i = 0; i < 4; i++)
{
grid.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(1, GridUnitType.Star)
});
}
for (int i = 0; i < 12; i++)
{
grid.RowDefinitions.Add(new RowDefinition {
Height = new GridLength(1, GridUnitType.Star)
});
}
// Add headlines to grid
grid.Children.Add(new Label {
Text = "Digit", Style = labelgridheadlineStyle
}, 0, 0);
grid.Children.Add(new Label {
Text = "Total", Style = labelgridheadlineStyle
}, 1, 0);
grid.Children.Add(new Label {
Text = "Correct", Style = labelgridheadlineStyle
}, 2, 0);
grid.Children.Add(new Label {
Text = "in %", Style = labelgridheadlineStyle
}, 3, 0);
// Add results for each digit to grid
for (int i = 0; i < 10; i++)
{
grid.Children.Add(new Label {
Text = i.ToString(), Style = labelgridStyle
}, 0, i + 1);
grid.Children.Add(new Label {
Text = String.Format("{0:N0}", digittotal[i]), Style = labelgridStyle
}, 1, i + 1);
grid.Children.Add(new Label {
Text = String.Format("{0:N0}", digitcorrect[i]), Style = labelgridStyle
}, 2, i + 1);
grid.Children.Add(new Label {
Text = String.Format("{0:P0}", digittotal[i] > 0 ? digitcorrect[i] / digittotal[i] : 0), Style = labelgridStyle
}, 3, i + 1);
}
// Add sum to grid
grid.Children.Add(new Label {
Text = "Sum", Style = labelgridhighlightStyle
}, 0, 11);
grid.Children.Add(new Label {
Text = String.Format("{0:N0}", digittotal.Sum()), Style = labelgridhighlightStyle
}, 1, 11);
grid.Children.Add(new Label {
Text = String.Format("{0:N0}", digitcorrect.Sum()), Style = labelgridhighlightStyle
}, 2, 11);
grid.Children.Add(new Label {
Text = String.Format("{0:P0}", digitcorrect.Sum() / digittotal.Sum()), Style = labelgridhighlightStyle
}, 3, 11);
// Define button for details
Button button = new Button
{
Text = "Details",
WidthRequest = Application.Current.MainPage.Width * 0.8,
HeightRequest = Application.Current.MainPage.Height * 0.10,
VerticalOptions = LayoutOptions.Center,
};
// Show page with detailed results for each digit
button.Clicked += (s, e) =>
{
if (messageflag == false)
{
DisplayAlert("Information", "Look at the details by swiping left and right.", "OK");
messageflag = true;
}
Navigation.PushAsync(new ResultsDetailsPage(neuralnet, mnistdata));
};
// Define page
Content = new ScrollView
{
Content = new StackLayout
{
Children = { grid, button },
Spacing = Application.Current.MainPage.Height * 0.05
}
};
}
/// <summary>
/// Copies the contents of the source neural network into this neural network.
/// </summary>
/// <param name="src">the source net, to be copied</param>
///
public void CopyNeuralNet(NeuralNet src)
{
mLayers.Clear();
mActivations.Clear();
mUnitInputs.Clear();
mActiveUnits.Clear();
mActiveSlope.Clear();
mActiveAmplify.Clear();
mNumInputs = src.mNumInputs;
mNumOutputs = src.mNumOutputs;
mNumLayers = src.mNumLayers;
mOutUnitType = src.mOutUnitType;
mOutUnitSlope = src.mOutUnitSlope;
mOutUnitAmplify = src.mOutUnitAmplify;
// the weighted connections linking the network layers
for (int i = 0; i < src.mLayers.Count; i++)
{
NNetWeightedConnect wCnct = new NNetWeightedConnect(src.mLayers[i]);
mLayers.Add(wCnct);
}
// the activation values for each of the network layers
int rowLen = src.mActivations.Count;
for (int row = 0; row < rowLen; row++)
{
List <double> vec = new List <double>();
int colLen = src.mActivations[row].Count;
for (int col = 0; col < colLen; col++)
{
vec.Add(src.mActivations[row][col]);
}
mActivations.Add(vec);
}
// the input values for the layer activation functions
rowLen = src.mUnitInputs.Count;
for (int row = 0; row < rowLen; row++)
{
List <double> vec = new List <double>();
int colLen = src.mUnitInputs[row].Count;
for (int col = 0; col < colLen; col++)
{
vec.Add(src.mUnitInputs[row][col]);
}
mUnitInputs.Add(vec);
}
// the hidden layer unit activation function types
for (int i = 0; i < src.mActiveUnits.Count; i++)
{
mActiveUnits.Add(src.mActiveUnits[i]);
}
// the hidden layer unit activation function slope values
for (int i = 0; i < src.mActiveSlope.Count; i++)
{
mActiveSlope.Add(src.mActiveSlope[i]);
}
// the hidden layer unit activation function amplify values
for (int i = 0; i < src.mActiveAmplify.Count; i++)
{
mActiveAmplify.Add(src.mActiveAmplify[i]);
}
}
public Neuron()
{
InputSynapses = new List <Synapse>(); //create an object from the input list
OutputSynapses = new List <Synapse>(); //create an object from the output list
Bias = NeuralNet.GetRandom(); //This value is randomized for every neuorn
}
public static Tensor4[] GetRandBatch(Tensor3[] x, int[] y, int batchSize, NeuralNet net)
{
Tensor4 batchX = new Tensor4(x[0].width, x[0].height, x[0].deep, batchSize);
Tensor4 batchY = new Tensor4(net.output.width, net.output.height, net.output.deep, batchSize);
int[] indexes = new int[batchSize];
int index = 0;
while (true)
{
indexes[index] = (int)(rand.NextDouble() * x.Length);
for (int i = 0; i < index; i++)
{
if (indexes[i] == indexes[index])
{
break;
}
if (i == index - 1)
{
index++;
break;
}
}
if (index == 0)
{
index++;
}
if (index == indexes.Length)
{
break;
}
}
for (int i = 0; i < batchSize; i++)
{
for (int zInd = 0; zInd < x[indexes[i]].deep; zInd++)
{
for (int yInd = 0; yInd < x[indexes[i]].height; yInd++)
{
for (int xInd = 0; xInd < x[indexes[i]].width; xInd++)
{
batchX[i, zInd, yInd, xInd] = x[indexes[i]][zInd, yInd, xInd];
}
}
}
Tensor3 Y = new Tensor3(net.output.width, net.output.height, net.output.deep);
if (net.output.width != 1)
{
Y[0, 0, y[indexes[i]]] = 1.0;
}
else
{
Y[y[indexes[i]], 0, 0] = 1.0;
}
for (int zInd = 0; zInd < Y.deep; zInd++)
{
for (int yInd = 0; yInd < Y.height; yInd++)
{
for (int xInd = 0; xInd < Y.width; xInd++)
{
batchY[i, zInd, yInd, xInd] = Y[zInd, yInd, xInd];
}
}
}
}
return(new Tensor4[] { batchX, batchY });
}
public ResultsCamPage(NeuralNet neuralnet, Vector <double> input, Vector <double> result)
{
// Define GUI
// Page properties
Title = "Results";
BackgroundColor = Color.SteelBlue;
Padding = new Thickness(Application.Current.MainPage.Width * 0.05, Application.Current.MainPage.Height * 0.05);
// Define styles for labels in grid
var labelgridStyle = new Style(typeof(Label))
{
Setters =
{
new Setter {
Property = Label.TextColorProperty, Value = Color.GhostWhite
},
new Setter {
Property = Label.BackgroundColorProperty, Value = Color.SteelBlue
},
new Setter {
Property = Label.HorizontalOptionsProperty, Value = LayoutOptions.FillAndExpand
},
new Setter {
Property = Label.HorizontalTextAlignmentProperty, Value = TextAlignment.Center
},
new Setter {
Property = Label.VerticalOptionsProperty, Value = LayoutOptions.FillAndExpand
},
new Setter {
Property = Label.VerticalTextAlignmentProperty, Value = TextAlignment.Center
}
}
};
var labelgridhighlightStyle = new Style(typeof(Label))
{
BasedOn = labelgridStyle,
Setters =
{
new Setter {
Property = Label.TextColorProperty, Value = Color.GhostWhite
},
new Setter {
Property = Label.BackgroundColorProperty, Value = Color.DarkOrange
},
new Setter {
Property = Label.FontAttributesProperty, Value = FontAttributes.Bold
}
}
};
var labelgridheadlineStyle = new Style(typeof(Label))
{
BasedOn = labelgridStyle,
Setters =
{
new Setter {
Property = Label.TextColorProperty, Value = Color.SteelBlue
},
new Setter {
Property = Label.BackgroundColorProperty, Value = Color.GhostWhite
},
new Setter {
Property = Label.FontAttributesProperty, Value = FontAttributes.Bold
}
}
};
// Define 11x3 grid for representing results
Grid grid = new Grid();
grid.BackgroundColor = Color.GhostWhite;
grid.ColumnSpacing = 1;
grid.RowSpacing = 1;
grid.Padding = new Thickness(1);
grid.HorizontalOptions = LayoutOptions.FillAndExpand;
grid.VerticalOptions = LayoutOptions.FillAndExpand;
for (int i = 0; i < 3; i++)
{
grid.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(1, GridUnitType.Star)
});
}
for (int i = 0; i < 11; i++)
{
grid.RowDefinitions.Add(new RowDefinition {
Height = new GridLength(1, GridUnitType.Star)
});
}
// Add headlines to grid
grid.Children.Add(new Label {
Text = "Digit", Style = labelgridheadlineStyle
}, 0, 0);
grid.Children.Add(new Label {
Text = "Ouput", Style = labelgridheadlineStyle
}, 1, 0);
grid.Children.Add(new Label {
Text = "Share", Style = labelgridheadlineStyle
}, 2, 0);
// Add content to grid
for (int i = 0; i < 10; i++)
{
if (result.AbsoluteMaximumIndex() == i)
{
grid.Children.Add(new Label {
Text = i.ToString(), Style = labelgridhighlightStyle
}, 0, i + 1);
grid.Children.Add(new Label {
Text = String.Format("{0:N3}", result[i]), Style = labelgridhighlightStyle
}, 1, i + 1);
grid.Children.Add(new Label {
Text = String.Format("{0:P1}", result[i] / result.Sum()), Style = labelgridhighlightStyle
}, 2, i + 1);
}
else
{
grid.Children.Add(new Label {
Text = i.ToString(), Style = labelgridStyle
}, 0, i + 1);
grid.Children.Add(new Label {
Text = String.Format("{0:N3}", result[i]), Style = labelgridStyle
}, 1, i + 1);
grid.Children.Add(new Label {
Text = String.Format("{0:P1}", result[i] / result.Sum()), Style = labelgridStyle
}, 2, i + 1);
}
}
// Define 2x3 grid for training net with own handwriting
Grid grid2 = new Grid();
grid2.BackgroundColor = Color.Transparent;
grid2.ColumnSpacing = 0;
grid2.RowSpacing = 0;
grid2.HorizontalOptions = LayoutOptions.Center;
grid2.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(8, GridUnitType.Star)
});
grid2.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(2, GridUnitType.Star)
});
grid2.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(1, GridUnitType.Star)
});
grid2.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(2, GridUnitType.Star)
});
grid2.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(1, GridUnitType.Star)
});
grid2.ColumnDefinitions.Add(new ColumnDefinition {
Width = new GridLength(2, GridUnitType.Star)
});
grid2.RowDefinitions.Add(new RowDefinition {
Height = new GridLength(1, GridUnitType.Star)
});
grid2.RowDefinitions.Add(new RowDefinition {
Height = new GridLength(1, GridUnitType.Star)
});
// Place content
grid2.Children.Add(new Label {
Text = "Net's answer:", Style = labelgridStyle, HorizontalTextAlignment = TextAlignment.Start
}, 0, 0);
grid2.Children.Add(new Label {
Text = "Correct answer:", Style = labelgridStyle, HorizontalTextAlignment = TextAlignment.Start
}, 0, 1);
grid2.Children.Add(new Label {
Text = result.AbsoluteMaximumIndex().ToString(), Style = labelgridStyle
}, 1, 0);
// labelresult keeps correct answer in Text property
Label labelresult = new Label {
Text = result.AbsoluteMaximumIndex().ToString(), Style = labelgridStyle
};
grid2.Children.Add(labelresult, 1, 1);
// Buttons for changing value of correct answer
Button buttondown = new Button()
{
Text = "-"
};
grid2.Children.Add(buttondown, 3, 4, 0, 2);
buttondown.Clicked += (s, e) =>
{
if (Convert.ToInt32(labelresult.Text) > 0)
{
labelresult.Text = (Convert.ToInt32(labelresult.Text) - 1).ToString();
}
else
{
labelresult.Text = "9";
}
};
Button buttonup = new Button()
{
Text = "+"
};
grid2.Children.Add(buttonup, 5, 6, 0, 2);
buttonup.Clicked += (s, e) =>
{
if (Convert.ToInt32(labelresult.Text) < 9)
{
labelresult.Text = (Convert.ToInt32(labelresult.Text) + 1).ToString();
}
else
{
labelresult.Text = "0";
}
};
// Define button to train net
Button button = new Button
{
Text = "Train this Digit",
WidthRequest = Application.Current.MainPage.Width * 0.8,
HeightRequest = Application.Current.MainPage.Height * 0.10,
VerticalOptions = LayoutOptions.Center,
};
// Train net
button.Clicked += (s, e) =>
{
// Determine output vector from correct answer
Vector <double> output = Vector <double> .Build.Dense(10, 0.01);
output[Convert.ToInt32(labelresult.Text)] = 0.99;
// Train net with this data set
neuralnet.Train(input, output);
// Increase number of training data sets used so far
neuralnet.TrainingDataCounter++;
DisplayAlert("Training completed", string.Format("Net trained with {0} digit{1} so far.", neuralnet.TrainingDataCounter, (neuralnet.TrainingDataCounter == 1 ? "" : "s")), "OK");
};
// Define page
Content = new ScrollView
{
Content = new StackLayout
{
Children = { grid, grid2, button },
Spacing = Application.Current.MainPage.Height * 0.05
}
};
}
public CameraTestPage(NeuralNet neuralnet)
{
// Save neural net
this.neuralnet = neuralnet;
// Define GUI
// Page properties
Title = "Handwriting";
BackgroundColor = Color.SteelBlue;
Padding = new Thickness(Application.Current.MainPage.Width * 0.05, Application.Current.MainPage.Height * 0.05);
// Define label
Label description = new Label
{
Text = "Take a picture of a single digit (0..9) with your phone camera or draw it directly with your finger on the screen"
};
// Define button to take pictures of digits
Button buttoncam = new Button
{
Text = "Take a Picture",
WidthRequest = Application.Current.MainPage.Width * 0.8,
HeightRequest = Application.Current.MainPage.Height * 0.10,
VerticalOptions = LayoutOptions.Center
};
buttoncam.Clicked += TakeAndShowPicture;
// Define button to draw a digit
Button buttondraw = new Button
{
Text = "Draw with Finger",
WidthRequest = Application.Current.MainPage.Width * 0.8,
HeightRequest = Application.Current.MainPage.Height * 0.10,
VerticalOptions = LayoutOptions.Center
};
buttondraw.Clicked += async(s, e) =>
{
// Open page to draw digit
// Note: DrawDigitPage writes result to "bitmap"
await Navigation.PushAsync(new DrawDigitPage());
};
// Define button to ask net to guess the digit
Button buttonasknet = new Button
{
Text = "Ask Neural Net",
WidthRequest = Application.Current.MainPage.Width * 0.8,
HeightRequest = Application.Current.MainPage.Height * 0.10,
VerticalOptions = LayoutOptions.Center,
};
buttonasknet.Clicked += AskNeuralNet;
// Define canvas for showing the picture of the digit
canvasview = new SKCanvasView
{
HorizontalOptions = LayoutOptions.FillAndExpand,
VerticalOptions = LayoutOptions.FillAndExpand
};
canvasview.PaintSurface += OnCanvasViewPaintSurface;
Content = new ScrollView
{
Content = new StackLayout
{
Children = { description, buttoncam, buttondraw, canvasview, buttonasknet },
Spacing = Application.Current.MainPage.Height * 0.05
}
};
// Convert bitmap and update canvas when page appears after drawing page
Appearing += (s, e) =>
{
if (bitmap == null)
{
return;
}
// Check if bitmap is not squared and not 28x28 px
if ((bitmap.Width != bitmap.Height) || bitmap.Width != 28)
{
// Convert bitmap to 28x28 pixel grayscale
bitmap = ConvertBitmap(bitmap);
// Update canvas
canvasview.InvalidateSurface();
}
};
}
public Synapse(Neuron inputNeuron, Neuron outputNeuron)
{
InputNeuron = inputNeuron;
OutputNeuron = outputNeuron;
Weight = NeuralNet.GetRandom();
}
public Neuron()
{
InputSynapses = new List <Synapse>();
OutputSynapses = new List <Synapse>();
Bias = NeuralNet.GetRandom();
}
public ResultsDetailsPage(NeuralNet neuralnet, MNISTDataManager mnistdata)
{
// Save parameters
this.neuralnet = neuralnet;
this.mnistdata = mnistdata;
// Define GUI
// Page properties
Title = "Details";
BackgroundColor = Color.SteelBlue;
Padding = new Thickness(Application.Current.MainPage.Width * 0.05, Application.Current.MainPage.Height * 0.05);
// Step 1: Generate initial collection of pictures from MNIST database (buffersize elements)
pictures = new ObservableCollection <CarouselItem>();
AddPictures(0);
// Step 2: Generate data template (just a Image with data binding to CarouselItem class)
DataTemplate template = new DataTemplate(() =>
{
Image image = new Image();
image.SetBinding(Image.SourceProperty, "Picture");
return(image);
});
// Step 3: Generate carousel view
var myCarousel = new CarouselViewControl();
myCarousel.Position = 0; //default
myCarousel.ItemsSource = pictures;
myCarousel.ItemTemplate = template;
myCarousel.InterPageSpacing = 2;
myCarousel.Orientation = CarouselViewOrientation.Horizontal;
myCarousel.ShowIndicators = false;
myCarousel.HorizontalOptions = LayoutOptions.FillAndExpand;
myCarousel.VerticalOptions = LayoutOptions.FillAndExpand;
// Define labels
labelheadline = new Label();
labeldigitcorrect = new Label();
labeldigitanswernet = new Label
{
HorizontalOptions = LayoutOptions.FillAndExpand,
HeightRequest = Application.Current.MainPage.Height * 0.1,
VerticalTextAlignment = TextAlignment.Center
};
SetLabels(0);
// Step 4: Build page
Content = new ScrollView
{
Content = new StackLayout
{
Children = { labelheadline, labeldigitcorrect, myCarousel, labeldigitanswernet },
Spacing = Application.Current.MainPage.Height * 0.05
}
};
// Update page when image is swiped
myCarousel.PositionSelected += (s, e) =>
{
// Add new pictures if user has reached end of carousel
// KNOWN ERROR: When end is reached the wrong picture is displayed (first picture of uploaded data)
// Warning: stepping through all MINST data may yield to memory problems
if (myCarousel.Position == pictures.Count - 1)
{
AddPictures(myCarousel.Position + 1);
}
// Update label
SetLabels(myCarousel.Position);
};
}
/// <summary>
/// Copy constructor.
/// </summary>
/// <param name="original">the NeuralNet object to be copied</param>
///
public NeuralNet(NeuralNet original)
{
CopyNeuralNet(original);
}
