public Brain(Brain model)
{
Neurons = model.Neurons;
Synapses = model.Synapses;
InputCount = model.InputCount;
Inputs = model.Inputs;
Expected = model.Expected;
}
public Configuration(Configuration template)
{
model = template.model;
SIAs = new List<Tuple<double, double>>();
SIAs.AddRange(template.SIAs);
Weights = new List<double>();
Weights.AddRange(template.Weights);
}
public Configuration(Brain model, bool randomize = false)
{
this.model = model;
SIAs = new List<Tuple<double, double>>();
Weights = new List<double>();
//assemble list of all neurons and then all synapses
foreach (List<Neuron> nlist in this.model.Neurons.Values)
foreach (Neuron n in nlist)
if (randomize)
SIAs.Add(new Tuple<double, double>(Neuron.GetRandomSlope(), Neuron.GetRandomAugment()));
else
SIAs.Add(new Tuple<double, double>(n.Slope, n.Augment));
foreach (List<Synapse> slist in this.model.Synapses.Values)
foreach (Synapse s in slist)
if (randomize)
Weights.Add(Synapse.GetRandomWeight());
else
Weights.Add(s.Weight);
}
public static Bitmap GetSchema(Brain brain)
{
int width = 1024;
int height = 768;
Font font;
int layernum = brain.Synapses.Count + 1;
int neuronsize = (int)((height * 0.8) / layernum / 2);
neuronsize = (neuronsize > height / 6) ? height / 6 : neuronsize;
String synapsesstr = brain.GetSynapsesStr();
// count neurons in each layer
List<int> neuroncount = new List<int>();
if (brain.Neurons != null)
{
neuroncount.Add(brain.InputCount);
foreach (List<Neuron> list in brain.Neurons.Values) // first already complete
neuroncount.Add(list.Count);
}
int max = neuroncount.Max();
// add missing synapses - should be fixed in Update()s ?
/*for (int i = 0; i < layernum - 1; i++) // for each layer
{
//g.DrawLine(Pens.White, 0, ycoords[layernum - i - 1], width, ycoords[layernum - i - 1]);
for (int j = 0; j < neuroncount[i]; j++) // for each neuron in layer
{
for (int k = 0; k < neuroncount[i + 1]; k++)
{
brain.layers[i].Neurons[k].AddMissingSynapse(j);
}
}
}*/
// compute y-coords
List<int> ycoords = new List<int>();
for (int i = 0; i < layernum; i++)
ycoords.Add((int)(height / (layernum) * i + height / (layernum + 3)));
if ((neuronsize + 2) * 1.5 * max > width) // neurons too big for width
neuronsize = (int)(width / max / 1.8);
font = new Font(new FontFamily("Arial"), (neuronsize / 4 > 0) ? neuronsize / 4 : 1, FontStyle.Bold, GraphicsUnit.Pixel);
Bitmap b = new Bitmap(width, height);
// prepare points
Dictionary<String, Point> coords = new Dictionary<String, Point>();
for (int i = 0; i < layernum; i++)
for (int j = 0; j < neuroncount[i]; j++)
coords.Add(String.Format("{0}_{1}", i, j),
// left margin, neuron spacing, from center
new Point((int)(neuronsize/2 + j * neuronsize * 1.5 + (width / 2) - ((neuronsize - 1) * 1.5 / 2 * neuroncount[i])),
ycoords[layernum - 1 - i]));
// draw
using (Graphics g = Graphics.FromImage(b))
{
g.Clear(Color.Black);
for (int i = 0; i < layernum; i++) // for each layer
{
//g.DrawLine(Pens.White, 0, ycoords[layernum - i - 1], width, ycoords[layernum - i - 1]);
// draw synapses
if (i != layernum - 1)
{
Pen p;
foreach (Synapse s in brain.Synapses[i - 1])
{
float weight = brain.maxweight == 0 ? 0 : (float)(s.Weight / brain.maxweight * neuronsize / 6);
if (weight == 0)
continue;
if (weight > 0)
p = new Pen(Brushes.White, weight);
else
p = new Pen(Brushes.Red, -weight);
String key = String.Format("{0}_{1}", i, s.Source == null ? s.InputIndex : s.Source.Index);
//Console.WriteLine(String.Format("{0} > {1}", key, String.Format("{0}_{1}", i + 1, s.Target.Index)));
g.DrawLine(p, coords[key], coords[String.Format("{0}_{1}", i + 1, s.Target.Index)]);
}
}
// draw the rest
for (int j = 0; j < neuroncount[i]; j++) // for each neuron in layer
{
string key = String.Format("{0}_{1}", i, j);
string caption;
if (i == 0) //input
caption = string.Format("i{0}", j);
else
caption = String.Format("n{0}_{1}", i - 1, j);
// draw circles
float neuronthickness = brain.maxaugment == 0 || i==0 ? 1 : Math.Abs((float)(brain.Neurons[i - 1][j].Augment/ brain.maxaugment * neuronsize / 12));
neuronthickness *=1.5f;
//neuronthickness += 2;
Pen neuronpen = new Pen((i == 0 || brain.Neurons[i - 1][j].Augment >= 0) ? Brushes.White : Brushes.Red, neuronthickness);
//g.FillEllipse((i==0 || brain.Neurons[i-1][j].Augment>=0)?Brushes.White:Brushes.Red, coords[key].X - neuronsize / 2, coords[key].Y - neuronsize / 2, neuronsize, neuronsize);
//g.FillEllipse(Brushes.Black, coords[key].X - neuronsize / 2+neuronthickness/2, coords[key].Y - neuronsize / 2+neuronthickness/2, neuronsize- neuronthickness, neuronsize- neuronthickness);
g.FillEllipse(Brushes.Black, coords[key].X - neuronsize / 2, coords[key].Y - neuronsize / 2, neuronsize, neuronsize);
g.DrawEllipse(neuronpen, coords[key].X - neuronsize / 2, coords[key].Y - neuronsize / 2, neuronsize, neuronsize);
// draw strings
SizeF stringsize = g.MeasureString(caption, font);
g.DrawString(caption, font, Brushes.White, coords[key].X - stringsize.Width / 2, coords[key].Y - stringsize.Height / 2);
}
}
}
return b;
}
private void btnNeuralRun_Click(object sender, EventArgs e)
{
ges = new Dictionary<int, double>(); // Global Errors
barNeuralMatch.Value = 0;
if (txtNeuralInput.Lines.Count((s) => s.Trim() != "") != txtNeuralExpected.Lines.Count((s) => s.Trim() != ""))
return;
// update NN
Update();
bool shouldloop = true;
//do the magic
for (int i = 0; i < (int)numNeuralEpoch.Value; i++)
{
switch (cmbNeuralAlgorithm.Text)
{
case "Activate": // basic computation
{
brain.Think();
shouldloop = false; // no reason to repeat
break;
}
case "SOMA": // basic computation
{
List<Neural.Configuration> population = new List<Neural.Configuration>();
for (int j = 0; j < 50; j++)
population.Add(new Neural.Configuration(brain, true));
// also use actual configuration
population.Add(new Neural.Configuration(brain));
population = new SOMA().Run(population);
//use the best one
population.Sort((x, y) => x.GE.CompareTo(y.GE));
brain.UpdateConfiguration(population[0]);
brain.Think();
break;
}
case "Fixed Increments":
{
brain.Think(NeuralNetworkAlgorithm.FixedIncrement);
brain.Think(); // check again (no adaptation) to update values
break;
}
case "Back Propagation":
{
brain = brain.Think(NeuralNetworkAlgorithm.BackPropagation);
brain.Think(); // to compute final outputs
break;
}
default:
{
txtLog.AppendText(String.Format("'{0}' algorithm is not implemented.\r\n", cmbNeuralAlgorithm.Text));
shouldloop = false;
break;
}
}
barNeuralProgress.Value = (i + 1) * 10000 / (int)numNeuralEpoch.Value;
txtNeuralOutput.Text = brain.GetDataStr(brain.Outputs);
barNeuralMatch.Value = (int)(brain.ComputeMatch() * 100);
UpdateNeuronDataGrid();
txtNeuralSynapses.Text = brain.GetSynapsesStr();
UpdateChartLSP();
ges.Add(i + 1, brain.GetGlobalError());
UpdateChartStatus();
InvalidateAll();
if (!shouldloop || brain.ComputeMatch() == 1)
{
double ge = brain.GetGlobalError();
for (int j = i + 1; j < numNeuralEpoch.Value; j++)
ges.Add(j + 1, ge);
barNeuralProgress.Value = barNeuralProgress.Maximum;
UpdateChartStatus();
break;
}
if (i != numNeuralEpoch.Value - 1)
{
int sleeptime = (numNeuralEpoch.Value > 10) ? (int)(1000 / numNeuralEpoch.Value) : 200;
if (sleeptime > 16) Thread.Sleep(sleeptime);
}
}
txtLog.AppendText(String.Format("Finished with Global Error of {0}.\n", brain.GetGlobalError()));
}
private void btnNeuralLoadConfiguration_Click(object sender, EventArgs e)
{
openFileDialog1.FileName = "network.ann";
if (openFileDialog1.ShowDialog() == DialogResult.OK)
{
IFormatter formatter = new BinaryFormatter();
Stream stream = new FileStream(openFileDialog1.FileName, FileMode.Open, FileAccess.Read, FileShare.Read);
brain = (Brain)formatter.Deserialize(stream);
stream.Close();
//show actual data
txtNeuralInput.Text = brain.GetDataStr(brain.Inputs);
txtNeuralExpected.Text = brain.GetDataStr(brain.Expected);
gridNeuralLayers.Rows.Clear();
for (int i = 0; i < brain.Neurons.Count; i++)
{
AddNewLayer(brain.Neurons[i].Count, functionlist.FirstOrDefault(x => x.Value.ToString() == brain.Neurons[i][0].f.ToString()).Key);
}
barNeuralMatch.Value = 0;
UpdateNeuronDataGrid();
txtNeuralSynapses.Text = brain.GetSynapsesStr();
}
}
public Brain Think(NeuralNetworkAlgorithm algo = NeuralNetworkAlgorithm.None)
{
if (algo != NeuralNetworkAlgorithm.BackPropagation)
{
if (Inputs == null)
return this;
Outputs = new List<List<double>>();
for (int inputcounter = 0; inputcounter < Inputs.Count; inputcounter++) // for every input set
{
ThinkOnce(inputcounter, algo);
}
return this;
}
else // back propagation
{
// http://mattmazur.com/2015/03/17/a-step-by-step-backpropagation-example/
double eta = 0.2;
//Outputs = new List<List<double>>();
Brain newbrain = new Brain(this);
for (int inputcounter = 0; inputcounter < Inputs.Count; inputcounter++) // for every input set
{
ThinkOnce(inputcounter);
// compute neuron errors - from the top, not for input neurons...
for (int i = newbrain.Neurons.Count - 1; i >= 0; i--)
{
for (int j = 0; j < newbrain.Neurons[i].Count; j++)
{
Neuron n = newbrain.Neurons[i][j];
if (i == newbrain.Neurons.Count - 1)
n.Error = Outputs[inputcounter][j] - Expected[inputcounter][j];
else
{
n.Error = 0;
foreach (Synapse s in newbrain.Synapses[i])
{
if (s.Source != n)
continue;
n.Error += s.Weight * s.Target.Error * n.f.ComputeDerivation(n.Input, n.Slope);
}
}
// and compute augments
double augdiff = eta * n.Error;
n.Augment -= augdiff;
}
}
// now update the weights
foreach (int i in newbrain.Synapses.Keys)
{
for (int j = 0; j < newbrain.Synapses[i].Count; j++)
{
Synapse s = newbrain.Synapses[i][j];
double weidiff = eta * s.Target.Error * s.LastInput;
weidiff+= s.LastDiff; // momentum-aware
s.LastDiff = weidiff;
s.Weight -= weidiff;
}
}
}
return newbrain;
}
}
