public static TeachingSample[] generateTeachingSamplesFromFile(string file)
{
List <TeachingSample> samples = new List <TeachingSample>();
try
{
StreamReader reader = File.OpenText(file);
string current_str;
while (!reader.EndOfStream)
{
current_str = reader.ReadLine();
string[] values = current_str.Split(':');
string input_str = values[0].Trim();
string output_str = values[1].Trim();
string[] input_strs = input_str.Split(',');
string[] output_strs = output_str.Split(',');
TeachingSample sample = new TeachingSample(input_strs.Length, output_strs.Length);
foreach (var input in input_strs)
{
sample.input_values[Array.IndexOf(input_strs, input)] = Convert.ToDouble(input);
}
foreach (var output in output_strs)
{
sample.desired_outputs[Array.IndexOf(output_strs, output)] = Convert.ToDouble(output);
}
samples.Add(sample);
}
}
catch (Exception)
{
return(null);
}
return(samples.ToArray());
}
private static void back_propagation(List <Neuron[]> ann_layers, double precision, TeachingSample teaching_sample, Teacher teacher)
{
Neuron[] input_layer = ann_layers.Find(layer => layer is InputNeuron[]);
Neuron[] hidden_layer = ann_layers.Find(layer => !(layer is InputNeuron[]) && !(layer is OutputNeuron[]));
Neuron[] output_layer = ann_layers.Find(layer => layer is OutputNeuron[]);
double[] deltas_output = new double[output_layer.Length];
double[] deltas_hidden = new double[hidden_layer.Length];
foreach (var neuron in output_layer)
{
double output = neuron.getAxonForTeacher(teacher).value;
double target = teaching_sample.desired_outputs[Array.IndexOf(output_layer, neuron)];
deltas_output[Array.IndexOf(output_layer, neuron)] = output * (1 - output) * (target - output);
}
double[] new_weights_output = new double[output_layer.Length * hidden_layer.Length];
int ind = 0;
foreach (var neuron in output_layer)
{
foreach (var synapse in neuron.getSynapsesForTeacher(teacher))
{
double last_weight = synapse.getWeightAsTeacher(teacher);
double new_weight = last_weight + (precision * deltas_output[Array.IndexOf(output_layer, neuron)] * synapse.axon.value);
new_weights_output[ind] = new_weight;
ind++;
}
}
foreach (var neuron in hidden_layer)
{
double output = neuron.getAxonForTeacher(teacher).value;
double sum = 0;
foreach (var out_neuron in output_layer)
{
double last_error = out_neuron.getSynapsesForTeacher(teacher).Find(syn => syn.axon == neuron.getAxonForTeacher(teacher)).getWeightAsTeacher(teacher);
sum += deltas_output[Array.IndexOf(output_layer, out_neuron)] * last_error;
}
deltas_hidden[Array.IndexOf(hidden_layer, neuron)] = output * (1 - output) * sum;
}
foreach (var neuron in hidden_layer)
{
foreach (var synapse in neuron.getSynapsesForTeacher(teacher))
{
double last_weight = synapse.getWeightAsTeacher(teacher);
double new_weight = last_weight + (precision * deltas_hidden[Array.IndexOf(hidden_layer, neuron)] * synapse.axon.value);
synapse.recalculateWeightAsTeacher(teacher, new_weight);
}
}
ind = 0;
foreach (var neuron in output_layer)
{
foreach (var synapse in neuron.getSynapsesForTeacher(teacher))
{
synapse.recalculateWeightAsTeacher(teacher, new_weights_output[ind]);
ind++;
}
}
}
private void bGo_Click(object sender, EventArgs e)
{
try
{
TeachingSample[] samples = null;
if (rbText.Checked)
{
MessageBox.Show("Please, choose text file to generate teaching samples");
OpenFileDialog dlg = new OpenFileDialog();
dlg.ShowDialog();
if (dlg.FileName != null && dlg.FileName != "")
{
samples = TeachingSample.generateTeachingSamplesFromFile(dlg.FileName);
if (samples == null)
{
throw new Exception();
}
}
}
else if (rbImage.Checked)
{
MessageBox.Show("Please, choose images to generate teaching samples");
OpenFileDialog dlg = new OpenFileDialog();
dlg.Multiselect = true;
dlg.ShowDialog();
if (dlg.FileNames.Length > 0)
{
samples = new TeachingSample[dlg.FileNames.Length];
for (int i = 0; i < samples.Length; i++)
{
List <string[]> anns_info = ann_manager.getTableInfo();
int input_length = 0;
int output_length = 0;
foreach (var info in anns_info)
{
if (info[0] == ann_manager.getSelectedANNName())
{
input_length = Convert.ToInt32(info[1]);
output_length = Convert.ToInt32(info[3]);
break;
}
}
samples[i] = new TeachingSample(input_length, output_length);
if (!samples[i].generateTeachingSampleFromImage(dlg.FileNames[i]))
{
MessageBox.Show("one of the files selected is bad");
return;
}
}
}
else
{
MessageBox.Show("No files for teaching sample selected!");
return;
}
}
else
{
MessageBox.Show("An odd situation occured. This is strange as f**k (also as div by 0), and program cant proceed...");
throw new Exception();
}
if (samples != null)
{
double precision = Convert.ToDouble(textBox2.Text);
int iterations = Convert.ToInt32(textBox1.Text);
Close();
string teachingResults = ann_manager.teachANN(precision, samples, iterations);
MessageBox.Show("Teaching results: " + teachingResults);
}
else
{
throw new Exception();
}
}
catch (Exception) { MessageBox.Show("An error occured. Probably, you entered wrong values, or choose wrong file(s)"); };
}
