/// <summary>
/// Start training, fade all controls
/// </summary>
private void ButtonStartClick(object sender, EventArgs e)
{
netTrainer = new NetTrainer(modelService);
_tbLearningRate.Enabled = false;
_tbMomentum.Enabled = false;
_buttonStart.Enabled = false;
_buttonPause.Enabled = true;
_buttonAbort.Enabled = true;
_cbLog.Enabled = false;
_tbHiddenUnits.Enabled = false;
_cmbActivationFunction.Enabled = false;
_cmbActivationFunctionHidden.Enabled = false;
_cmbActivationFunctionOutput.Enabled = false;
startTime = DateTime.Now;
_tElapsedTime.Enabled = true;
_tElapsedTime.Start();
netTrainer.StartTrainingAsync(netToTrain, CallBack);
}
/// <summary>
/// Start button clicked. Assembling started.
/// </summary>
private void BtnStartClick(object sender, EventArgs e)
{
Invoke(new Action(() =>
{
_btnStart.Enabled = false;
_btnInsert.Enabled = false;
_nudNumberOfGroupsNeuralHasher.Enabled = false;
_nudNumberOfHashesPerKeyNeuralHasher.Enabled = false;
_btnSaveEnsamble.Enabled = false;
_tbSaveToEnsembleFilename.ReadOnly = true;
}));
/*Ensemble*/
Action action = () =>
{
List<Network> networks = new List<Network>();
/*Load all the networks*/
foreach (KeyValuePair<string, bool> item in dictionaryPathToNetworks)
{
if (item.Value)
{
networks.Add((Network)SerializeObject.Load(item.Key));
if (networks[networks.Count - 1].MedianResponces == null) networks.RemoveAt(networks.Count - 1);
}
}
if ((networks.Count*10 /*Number of network outputs*/) < (numberofgroupsneuralhasher*numberofhashesperkeyneuralhasher))
{
MessageBox.Show(string.Format("Not enough networks to create an ensemble of such size {0} {1}", numberofgroupsneuralhasher, numberofhashesperkeyneuralhasher));
return;
}
/*Load a network trainer*/
IActivationFunction function = networks[0].GetActivation(0);
NetTrainer netTrainer = new NetTrainer(modelService);
double[][] inputs = null, outputs = null; /*Normalized input/output pairs*/
Dictionary<Int32, List<BasicMLData>> trackIdFingerprints = netTrainer.GetNormalizedTrackFingerprints(function, 10, 10);
double[][] binaryCodes = netTrainer.GetNormalizedBinaryCodes(function, 10);
Tuple<double[][], double[][]> tuple = netTrainer.FillStandardInputsOutputs(trackIdFingerprints, binaryCodes);
inputs = tuple.Item1;
outputs = tuple.Item2;
/*Construct outputs using median response*/
int samplesCount = inputs.Length; /*10240*/
NeuralNetworkEnsemble nNEnsembe = new NeuralNetworkEnsemble(networks.ToArray()); /*40 networks*/
for (int i = 0; i < samplesCount /*1024 * 10*/; i++)
{
byte[] outputVec = nNEnsembe.ComputeHash(inputs[i]); /*Hash the inputs, returns 10*40 = 400*/
outputs[i] = new double[outputVec.Length]; /*400*/
for (int j = 0; j < outputVec.Length; j++)
{
outputs[i][j] = outputVec[j]; /*10240x400 matrix*/
}
}
/*At this point we have a the 10240 hash vectors [0, 400] which represent the outputs for the actual input*/
/*Calculate minimal mutual information between those outputs*/
MinimalMutualInfoPattern mmiPattern = new MinimalMutualInfoPattern(numberofgroupsneuralhasher, numberofhashesperkeyneuralhasher);
mmiPattern.CreatePattern(outputs);
nNEnsembe.HashPattern = mmiPattern;
nNEnsembe.Save(pathToEnsemble);
};
action.BeginInvoke((result) =>
{
/*Ensemble ended*/
action.EndInvoke(result);
Invoke(new Action(() =>
{
_btnStart.Enabled = true;
_btnInsert.Enabled = true;
_nudNumberOfGroupsNeuralHasher.Enabled = true;
_nudNumberOfHashesPerKeyNeuralHasher.Enabled = true;
_btnSaveEnsamble.Enabled = true;
_tbSaveToEnsembleFilename.ReadOnly = false;
}));
MessageBox.Show(Resources.EnsemblingEnded, Resources.Success, MessageBoxButtons.OK, MessageBoxIcon.Information);
}, action);
}