private Action _AddText(string s)
{
return(() =>
{
var lines = _textArea.Text.Split(new[] { "\n" }, StringSplitOptions.None);
if (lines.Length < 16)
{
_textArea.Append($"{s}\n");
}
else
{
_textArea.Text = "";
for (var i = 1; i < 16; i++)
{
lines[i - 1] = lines[i];
}
foreach (var line in lines)
{
if (line != "")
{
_textArea.Append($"{line}\n");
}
}
_textArea.Append($"{s}\n");
}
});
}
/// <summary>
/// Posts the log async. This prevents the UI from hanging if there's a high volume of logs.
/// </summary>
private static void PostLogAsync(object source, System.Timers.ElapsedEventArgs e)
{
//run always on UI thread
Application.Instance.Invoke(() =>
{
if (buffer.Length > 0)
{
outputArea.Append(buffer, true);
buffer = "";
}
timer.AutoReset = false;
});
}
public MainView()
{
Content = BuildContent();
ConfigureEvents();
ConfigureDataBindings();
this.Load += (_, e) => _textLog.Focus();
this.DataContextChanged += (_, e) =>
{
if (this.DataContext is MainViewModel vm)
{
vm.LogReceived += (__, ev) =>
{
Application.Instance.Invoke(() =>
{
_textLog.Append($"[{ev.TimeStamp.ToString()}] {ev.LogLevel} {ev.Message.TrimEnd('\r', '\n')}{Environment.NewLine}", true);
});
};
}
};
}
public MainForm()
{
Title = "Instant Calculator";
ClientSize = new Size(400, 350);
// Main content.
// 1. Input text area
// 2. Button to calculate
// 3. Output text area (read only)
Content = new StackLayout
{
Padding = 10,
Spacing = 10,
Orientation = Orientation.Vertical,
HorizontalContentAlignment = HorizontalAlignment.Stretch,
VerticalContentAlignment = VerticalAlignment.Stretch,
Items =
{
new StackLayoutItem
{
Control = _textInput = new TextArea
{
SpellCheck = false,
Wrap = true,
ReadOnly = false
},
Expand = false,
},
new StackLayout
{
Padding = 0,
Spacing = 5,
Orientation = Orientation.Horizontal,
VerticalContentAlignment = VerticalAlignment.Center,
HorizontalContentAlignment = HorizontalAlignment.Stretch,
Items =
{
(_buttonCalculate = new Button{
Text = "Calculate"
}),
new StackLayoutItem
{
Control = _textOutput = new TextBox
{
ReadOnly = true
},
Expand = true,
},
},
},
new StackLayoutItem
{
Control = _textLog = new TextArea
{
SpellCheck = false,
Wrap = true,
ReadOnly = true
},
Expand = true,
},
}
};
// Quit application
var quitCommand = new Command
{
MenuText = "Quit",
Shortcut = Application.Instance.CommonModifier | Keys.Q
};
quitCommand.Executed += (_, e) => Application.Instance.Quit();
// Show about dialog
var aboutCommand = new Command
{
MenuText = "About..."
};
aboutCommand.Executed +=
(_, e) => new AboutDialog().ShowDialog(this);
// Create menu
Menu = new MenuBar
{
Items = { },
ApplicationItems = { },
QuitItem = quitCommand,
AboutItem = aboutCommand
};
// Create toolbar, empty for now
ToolBar = new ToolBar {
Items = { }
};
DataContextChanged += (_, e) =>
{
if (!(this.DataContext is MainViewModel vm))
{
return;
}
vm.AppendLogAction = (log) => _textLog.Append(log, true);
ConfigureDataBinding();
};
DataContext = new MainViewModel();
}
public MainForm(Cpu cpu, int scale)
{
_cpu = cpu;
var dia = new OpenFileDialog {
MultiSelect = false
};
var result = dia.ShowDialog(this);
if (result == DialogResult.Ok || result == DialogResult.Yes)
{
_cpu.LoadRom(dia.FileName);
}
else
{
return;
}
var scaleFactor = scale;
Title = "StonerAte";
ClientSize = new Size(1000, 600);
_drawable.Size = new Size(64 * scaleFactor, 32 * scaleFactor);
_drawable.Paint += (sender, e) =>
{
var black = new Color(0, 0, 0);
var white = new Color(255, 255, 255);
for (var x = 0; x < _cpu.Gfx.GetLength(0); x++)
{
for (var y = 0; y < _cpu.Gfx.GetLength(1); y++)
{
e.Graphics.FillRectangle(_cpu.Gfx[x, y] == 1 ? white : black, x * scaleFactor, y * scaleFactor,
scaleFactor, scaleFactor);
}
}
};
_textArea.ReadOnly = true;
_textArea.Size = new Size(1000 - ((64 * scaleFactor) + 30), 32 * scaleFactor);
var basicsLayout = new DynamicLayout();
basicsLayout.BeginVertical();
_basicsLabels[0] = new Label {
Text = "Opcode:"
};
_basicsBoxs[0] = new TextBox {
ReadOnly = true
};
_basicsLabels[1] = new Label {
Text = "I:"
};
_basicsBoxs[1] = new TextBox {
ReadOnly = true
};
_basicsLabels[2] = new Label {
Text = "PC:"
};
_basicsBoxs[2] = new TextBox {
ReadOnly = true
};
_basicsLabels[3] = new Label {
Text = "SP:"
};
_basicsBoxs[3] = new TextBox {
ReadOnly = true
};
_basicsLabels[4] = new Label {
Text = "DrawFlag:"
};
_basicsBoxs[4] = new TextBox {
ReadOnly = true
};
for (var i = 0; i < 5; i++)
{
basicsLayout.BeginHorizontal();
basicsLayout.Add(_basicsLabels[i]);
basicsLayout.Add(_basicsBoxs[i]);
basicsLayout.EndHorizontal();
}
basicsLayout.BeginHorizontal();
basicsLayout.Add(_sTLabel);
basicsLayout.Add(_sTBox);
basicsLayout.EndHorizontal();
basicsLayout.BeginHorizontal();
basicsLayout.Add(_dTLabel);
basicsLayout.Add(_dTBox);
basicsLayout.EndHorizontal();
basicsLayout.EndVertical();
var basicsBox = new GroupBox
{
Text = "Basics",
Content = basicsLayout
};
var regLayout = new DynamicLayout();
regLayout.BeginVertical();
for (var i = 0; i < 8; i++)
{
_regBoxs[i] = new TextBox {
ReadOnly = true
};
_regLabels[i] = new Label {
Text = $"V[{i:X}]"
};
_regBoxs[i + 8] = new TextBox {
ReadOnly = true
};
_regLabels[i + 8] = new Label {
Text = $"V[{i+8:X}]"
};
regLayout.BeginHorizontal();
regLayout.Add(_regLabels[i]);
regLayout.Add(_regBoxs[i]);
regLayout.Add(_regLabels[i + 8]);
regLayout.Add(_regBoxs[i + 8]);
regLayout.EndHorizontal();
}
regLayout.EndVertical();
var registerBox = new GroupBox
{
Text = "Registers",
Content = regLayout
};
var controlLayout = new DynamicLayout();
controlLayout.BeginVertical();
controlLayout.BeginHorizontal();
controlLayout.Add(new Label
{
Text = "Freq: ",
});
controlLayout.Add(_freqBox);
controlLayout.EndHorizontal();
controlLayout.BeginHorizontal();
var startButton = new Button {
Text = "Start"
};
startButton.Click += (sender, args) => { _isRunning = true; };
var pauseButton = new Button {
Text = "Pause"
};
pauseButton.Click += (sender, args) => { _isRunning = false; };
controlLayout.Add(startButton);
controlLayout.Add(pauseButton);
controlLayout.EndHorizontal();
controlLayout.EndVertical();
var controlBox = new GroupBox
{
Text = "Controls",
Content = controlLayout,
};
var rom = new TextArea {
Size = new Size(200, 500)
};
for (var i = 0x200; i < 4096; i++)
{
rom.Append($"{i:X4} - {_cpu.Memory[i]:X2}\n");
}
var layout = new PixelLayout();
layout.Add(_drawable, 10, 10);
layout.Add(_textArea, (64 * scaleFactor) + 20, 10);
layout.Add(basicsBox, 10, (32 * scaleFactor) + 20);
layout.Add(registerBox, 260, (32 * scaleFactor) + 20);
layout.Add(controlBox, 10, (32 * scaleFactor) + 240);
layout.Add(rom, 750, (32 * scaleFactor) + 20);
Content = layout;
KeyDown += (sender, args) => { _cpu.Key = args; };
LoadComplete += (sender, args) =>
{
var thread = new Thread(RunCpu);
thread.Start();
};
}
protected override void OnLoad(EventArgs e)
{
base.OnLoad(e);
text.Append(string.Empty, true);
}
public void Train(IFlowsheet flowsheet, TextArea ta = null, Eto.OxyPlot.Plot plot = null)
{
var nl = Environment.NewLine;
var g = tf.Graph();
g.as_default();
if (session != null)
{
session.Dispose(); session = null;
}
session = tf.Session(graph: g);
tf_with(tf.variable_scope("Train"), delegate
{
if (flowsheet != null)
{
flowsheet.ShowMessage("Training Started...", IFlowsheet.MessageType.Information);
}
else
{
Application.Instance.Invoke(() =>
{
ta.Append("Training Started..." + nl, true);
});
}
// tf Graph Input
var X = tf.placeholder(tf.float32, shape: (-1, n_x), name: "X");
var Y = tf.placeholder(tf.float32, shape: (-1, n_y), name: "Y");
Tensor outlayer = null;
var sigma = 1.0f;
var weight_initializer = tf.variance_scaling_initializer(mode: "FAN_AVG", uniform: true, factor: sigma);
var bias_initializer = tf.zeros_initializer;
var n_neurons_1 = Parameters.NumberOfNeuronsOnFirstLayer;
var n_neurons_2 = n_neurons_1 / 2;
var n_neurons_3 = n_neurons_2 / 2;
var n_neurons_4 = n_neurons_3 / 2;
RefVariable W_hidden_1, W_hidden_2, W_hidden_3, W_hidden_4, W_out;
RefVariable bias_hidden_1, bias_hidden_2, bias_hidden_3, bias_hidden_4, bias_out;
Tensor hidden_1, hidden_2, hidden_3, hidden_4;
switch (Parameters.NumberOfLayers)
{
case 2:
// Hidden weights
W_hidden_1 = tf.Variable(weight_initializer.call(new int[] { n_x, n_neurons_1 }, dtype: TF_DataType.TF_FLOAT), name: "W1");
bias_hidden_1 = tf.Variable(bias_initializer.call(n_neurons_1, dtype: TF_DataType.TF_FLOAT), name: "b1");
W_hidden_2 = tf.Variable(weight_initializer.call(new int[] { n_neurons_1, n_neurons_2 }, dtype: TF_DataType.TF_FLOAT), name: "W2");
bias_hidden_2 = tf.Variable(bias_initializer.call(n_neurons_2, dtype: TF_DataType.TF_FLOAT), name: "b2");
// Output weights
W_out = tf.Variable(weight_initializer.call(new int[] { n_neurons_2, n_y }, dtype: TF_DataType.TF_FLOAT), name: "Wout");
bias_out = tf.Variable(bias_initializer.call(n_y, dtype: TF_DataType.TF_FLOAT), name: "bout");
// Hidden layer
hidden_1 = tf.nn.relu(tf.add(tf.matmul(X, W_hidden_1), bias_hidden_1), name: "h1");
hidden_2 = tf.nn.relu(tf.add(tf.matmul(hidden_1, W_hidden_2), bias_hidden_2), name: "h2");
// Output layer
outlayer = tf.add(tf.matmul(hidden_2, W_out), bias_out, name: "out");
break;
case 3:
// Hidden weights
W_hidden_1 = tf.Variable(weight_initializer.call(new int[] { n_x, n_neurons_1 }, dtype: TF_DataType.TF_FLOAT), name: "W1");
bias_hidden_1 = tf.Variable(bias_initializer.call(n_neurons_1, dtype: TF_DataType.TF_FLOAT), name: "b1");
W_hidden_2 = tf.Variable(weight_initializer.call(new int[] { n_neurons_1, n_neurons_2 }, dtype: TF_DataType.TF_FLOAT), name: "W2");
bias_hidden_2 = tf.Variable(bias_initializer.call(n_neurons_2, dtype: TF_DataType.TF_FLOAT), name: "b2");
W_hidden_3 = tf.Variable(weight_initializer.call(new int[] { n_neurons_2, n_neurons_3 }, dtype: TF_DataType.TF_FLOAT), name: "W3");
bias_hidden_3 = tf.Variable(bias_initializer.call(n_neurons_3, dtype: TF_DataType.TF_FLOAT), name: "b3");
// Output weights
W_out = tf.Variable(weight_initializer.call(new int[] { n_neurons_3, n_y }, dtype: TF_DataType.TF_FLOAT), name: "Wout");
bias_out = tf.Variable(bias_initializer.call(n_y, dtype: TF_DataType.TF_FLOAT), name: "bout");
// Hidden layer
hidden_1 = tf.nn.relu(tf.add(tf.matmul(X, W_hidden_1), bias_hidden_1), name: "h1");
hidden_2 = tf.nn.relu(tf.add(tf.matmul(hidden_1, W_hidden_2), bias_hidden_2), name: "h2");
hidden_3 = tf.nn.relu(tf.add(tf.matmul(hidden_2, W_hidden_3), bias_hidden_3), name: "h3");
// Output layer
outlayer = tf.add(tf.matmul(hidden_3, W_out), bias_out, name: "out");
break;
case 4:
// Hidden weights
W_hidden_1 = tf.Variable(weight_initializer.call(new int[] { n_x, n_neurons_1 }, dtype: TF_DataType.TF_FLOAT), name: "W1");
bias_hidden_1 = tf.Variable(bias_initializer.call(n_neurons_1, dtype: TF_DataType.TF_FLOAT), name: "b1");
W_hidden_2 = tf.Variable(weight_initializer.call(new int[] { n_neurons_1, n_neurons_2 }, dtype: TF_DataType.TF_FLOAT), name: "W2");
bias_hidden_2 = tf.Variable(bias_initializer.call(n_neurons_2, dtype: TF_DataType.TF_FLOAT), name: "b2");
W_hidden_3 = tf.Variable(weight_initializer.call(new int[] { n_neurons_2, n_neurons_3 }, dtype: TF_DataType.TF_FLOAT), name: "W3");
bias_hidden_3 = tf.Variable(bias_initializer.call(n_neurons_3, dtype: TF_DataType.TF_FLOAT), name: "b3");
W_hidden_4 = tf.Variable(weight_initializer.call(new int[] { n_neurons_3, n_neurons_4 }, dtype: TF_DataType.TF_FLOAT), name: "W4");
bias_hidden_4 = tf.Variable(bias_initializer.call(n_neurons_4, dtype: TF_DataType.TF_FLOAT), name: "b4");
// Output weights
W_out = tf.Variable(weight_initializer.call(new int[] { n_neurons_4, n_y }, dtype: TF_DataType.TF_FLOAT), name: "Wout");
bias_out = tf.Variable(bias_initializer.call(n_y, dtype: TF_DataType.TF_FLOAT), name: "bout");
// Hidden layer
hidden_1 = tf.nn.relu(tf.add(tf.matmul(X, W_hidden_1), bias_hidden_1), name: "h1");
hidden_2 = tf.nn.relu(tf.add(tf.matmul(hidden_1, W_hidden_2), bias_hidden_2), name: "h2");
hidden_3 = tf.nn.relu(tf.add(tf.matmul(hidden_2, W_hidden_3), bias_hidden_3), name: "h3");
hidden_4 = tf.nn.relu(tf.add(tf.matmul(hidden_3, W_hidden_4), bias_hidden_4), name: "h4");
// Output layer
outlayer = tf.add(tf.matmul(hidden_4, W_out), bias_out, name: "out");
break;
}
// Mean squared error
var mse = tf.reduce_sum(tf.pow(outlayer - Y, 2.0f), name: "mse");
var learn_rate = tf.constant(Parameters.LearningRate);
var opt = tf.train.AdamOptimizer(learn_rate).minimize(mse);
// Fit neural net
var batch_size = Parameters.BatchSize;
var mse_train = new List <float>();
var mse_test = new List <float>();
// Initialize the variables (i.e. assign their default value)
var init = tf.global_variables_initializer();
// Run the initializer
session.run(init);
// Start training
var epochs = Parameters.NumberOfEpochs;
foreach (var e in range(epochs))
{
// Shuffle training data
var shuffle_indices = np.random.permutation(np.arange(len(x_train)));
var shuffled_x = new NDArray(np.float32, x_train.shape);
var shuffled_y = new NDArray(np.float32, y_train.shape);
int i0 = 0;
foreach (var idx0 in shuffle_indices)
{
shuffled_x[i0] = x_train[idx0];
shuffled_y[i0] = y_train[idx0];
i0 += 1;
}
// Minibatch training
foreach (var i in range(0, len(y_train) / batch_size))
{
var start = i * batch_size;
var batch_x = shuffled_x[start.ToString() + ":" + (start + batch_size).ToString(), Slice.All];
var batch_y = shuffled_y[start.ToString() + ":" + (start + batch_size).ToString(), Slice.All];
// Run optimizer with batch
session.run(opt, (X, batch_x), (Y, batch_y));
// Show progress
var divrem = 0;
Math.DivRem(e, 5, out divrem);
if (divrem == 0)
{
// MSE train and test
mse_train.Add(session.run(mse, (X, x_train), (Y, y_train)));
mse_test.Add(session.run(mse, (X, x_test), (Y, y_test)));
if (flowsheet != null)
{
flowsheet.ShowMessage("Epoch: " + e.ToString(), IFlowsheet.MessageType.Information);
flowsheet.ShowMessage("MSE (training): " + mse_train.Last().ToString(), IFlowsheet.MessageType.Information);
flowsheet.ShowMessage("MSE (testing): " + mse_test.Last().ToString(), IFlowsheet.MessageType.Information);
}
else
{
Application.Instance.Invoke(() =>
{
ta.Append("Epoch: " + e.ToString() + nl, true);
ta.Append("MSE (training): " + mse_train.Last().ToString() + nl, true);
ta.Append("MSE (testing): " + mse_test.Last().ToString() + nl, true);
(plot.Model.Series[0] as OxyPlot.Series.LineSeries).Points.Add(new DataPoint(e, mse_train.Last()));
(plot.Model.Series[1] as OxyPlot.Series.LineSeries).Points.Add(new DataPoint(e, mse_test.Last()));
plot.Model.InvalidatePlot(true);
});
}
if (e > 10 &&
(Math.Abs(mse_train.Last() - mse_train[mse_train.Count - 2]) / mse_train[mse_train.Count - 2] <
Parameters.RelativeMSETolerance))
{
break;
}
}
}
}
if (flowsheet != null)
{
flowsheet.ShowMessage("Training Finished!", IFlowsheet.MessageType.Information);
}
else
{
Application.Instance.Invoke(() =>
{
ta.Append("Training Finished!" + nl, true);
});
}
x_test_unscaled = new NDArray(np.float32, x_test.shape);
x_train_unscaled = new NDArray(np.float32, x_train.shape);
for (var i = 0; i < x_test.shape[0]; i++)
{
for (var j = 0; j < x_test.shape[1]; j++)
{
x_test_unscaled[i][j] = Classes.Utils.UnScale(x_test[i][j],
Parameters.MinValues[j],
Parameters.MaxValues[j],
Parameters.MinScale,
Parameters.MaxScale);
}
}
for (var i = 0; i < x_train.shape[0]; i++)
{
for (var j = 0; j < x_train.shape[1]; j++)
{
x_train_unscaled[i][j] = Classes.Utils.UnScale(x_train[i][j],
Parameters.MinValues[j],
Parameters.MaxValues[j],
Parameters.MinScale,
Parameters.MaxScale);
}
}
var idx = Parameters.Labels.IndexOf(Parameters.Labels_Outputs.First());
y_test_unscaled = new NDArray(np.float32, y_test.shape);
y_train_unscaled = new NDArray(np.float32, y_train.shape);
for (var i = 0; i < y_test.shape[0]; i++)
{
for (var j = 0; j < y_test.shape[1]; j++)
{
y_test_unscaled[i][j] = Classes.Utils.UnScale(y_test[i][j],
Parameters.MinValues[idx + j],
Parameters.MaxValues[idx + j],
Parameters.MinScale,
Parameters.MaxScale);
}
}
for (var i = 0; i < y_train.shape[0]; i++)
{
for (var j = 0; j < y_train.shape[1]; j++)
{
y_train_unscaled[i][j] = Classes.Utils.UnScale(y_train[i][j],
Parameters.MinValues[idx + j],
Parameters.MaxValues[idx + j],
Parameters.MinScale,
Parameters.MaxScale);
}
}
yp_test = session.run(outlayer, (X, x_test));
yp_train = session.run(outlayer, (X, x_train));
yp_test_unscaled = new NDArray(np.float32, yp_test.shape);
yp_train_unscaled = new NDArray(np.float32, yp_train.shape);
for (var i = 0; i < yp_test.shape[0]; i++)
{
for (var j = 0; j < yp_test.shape[1]; j++)
{
yp_test_unscaled[i][j] = Classes.Utils.UnScale(yp_test[i][j],
Parameters.MinValues[idx + j],
Parameters.MaxValues[idx + j],
Parameters.MinScale,
Parameters.MaxScale);
}
}
for (var i = 0; i < yp_train.shape[0]; i++)
{
for (var j = 0; j < yp_train.shape[1]; j++)
{
yp_train_unscaled[i][j] = Classes.Utils.UnScale(yp_train[i][j],
Parameters.MinValues[idx + j],
Parameters.MaxValues[idx + j],
Parameters.MinScale,
Parameters.MaxScale);
}
}
// Testing example
var training_cost = session.run(mse, (X, x_train), (Y, y_train));
var testing_cost = session.run(mse, (X, x_test), (Y, y_test));
var diff = Math.Abs((float)training_cost - (float)testing_cost);
if (flowsheet != null)
{
flowsheet.ShowMessage($"Training Cost = {testing_cost}", IFlowsheet.MessageType.Information);
flowsheet.ShowMessage($"Testing Cost = {testing_cost}", IFlowsheet.MessageType.Information);
flowsheet.ShowMessage($"Absolute MSE = {diff}", IFlowsheet.MessageType.Information);
}
else
{
Application.Instance.Invoke(() =>
{
ta.Append($"Training Cost = {testing_cost}" + nl, true);
ta.Append($"Testing Cost = {testing_cost}" + nl, true);
ta.Append($"Absolute MSE = {diff}" + nl, true);
});
}
});
