private void m_solver_OnTrainingIteration(object sender, TrainingIterationArgs <T> e)
{
if (m_sw.Elapsed.TotalMilliseconds > 1000)
{
m_dfLastLoss = e.SmoothedLoss;
m_dfLastLearningRate = e.LearningRate;
updateStatus(e.Iteration, m_solver.MaximumIteration, e.Accuracy, e.SmoothedLoss, e.LearningRate);
m_sw.Restart();
}
}
/// <summary>
/// Event called on each training iterations.
/// </summary>
/// <param name="sender">Specifies who sent the event.</param>
/// <param name="e">Specifies the event parameters.</param>
private void m_mycaffe_OnTrainingIteration(object sender, TrainingIterationArgs <float> e)
{
if (m_swTraining.Elapsed.TotalMilliseconds > 1000)
{
Console.WriteLine("Iteration = " + e.Iteration.ToString("N0") + " Loss = " + e.SmoothedLoss.ToString());
m_swTraining.Restart();
}
if (e.Iteration % 30 == 0)
{
m_plots.Add(e.Iteration, e.SmoothedLoss);
}
}
/// <summary>
/// Called on each training iteration of the input model used to detect each hand written character.
/// </summary>
/// <param name="sender">Specifies the event sender.</param>
/// <param name="e">Specifies the event args.</param>
private void m_mycaffeInput_OnTrainingIteration(object sender, TrainingIterationArgs <float> e)
{
if (m_sw.Elapsed.TotalMilliseconds > 1000)
{
m_log.Progress = e.Iteration / (double)m_model.Iterations;
m_log.WriteLine("MNIST Iteration " + e.Iteration.ToString() + " of " + m_model.Iterations.ToString() + ", loss = " + e.SmoothedLoss.ToString());
m_sw.Restart();
m_plotsInputLoss.Add(e.Iteration, e.SmoothedLoss);
Image img = SimpleGraphingControl.QuickRender(m_plotsInputLoss, pbImage.Width, pbImage.Height, false, null, null, true, m_rgZeroLine);
m_bw.ReportProgress(1, img);
}
}
private void Solver_OnTrainingIteration(object sender, TrainingIterationArgs <T> e)
{
m_log.WriteLine("Iteration = " + e.Iteration.ToString() + " - Loss = " + e.SmoothedLoss.ToString());
if (double.IsNaN(e.Loss))
{
m_log.WriteError(new Exception("Loss = NAN!"));
m_evtCancel.Set();
return;
}
if (double.IsInfinity(e.Loss))
{
m_log.WriteError(new Exception("Loss = Infinity!"));
m_evtCancel.Set();
return;
}
}
/// <summary>
/// Called on each training iteration of the sequence model used to encode each detected hand written character
/// and then decode the encoding into the proper section of the Sin curve.
/// </summary>
/// <param name="sender">Specifies the event sender.</param>
/// <param name="e">Specifies the event args.</param>
private void m_mycaffe_OnTrainingIteration(object sender, TrainingIterationArgs <float> e)
{
if (m_sw.Elapsed.TotalMilliseconds > 1000)
{
m_log.Progress = e.Iteration / (double)m_model.Iterations;
m_log.WriteLine("Seq2Seq Epoch " + m_nTotalEpochs.ToString() + " Sequence " + m_nTotalSequences.ToString() + " Iteration " + e.Iteration.ToString() + " of " + m_model.Iterations.ToString() + ", loss = " + e.SmoothedLoss.ToString(), true);
m_sw.Restart();
m_fTotalCost += (float)e.SmoothedLoss;
m_nTotalIter1++;
float fLoss = m_fTotalCost / m_nTotalIter1;
m_plotsSequenceLoss.Add(m_nTotalSequences, fLoss);
if (m_plotsSequenceLoss.Count > 2000)
{
m_plotsSequenceLoss.RemoveAt(0);
}
Image img = SimpleGraphingControl.QuickRender(m_plotsSequenceLoss, pbImageLoss.Width, pbImageLoss.Height, false, null, null, true, m_rgZeroLine);
m_bw.ReportProgress(1, new Tuple <Image, int>(img, 0));
}
}
private void ctrl_OnTrainingIteration(object sender, TrainingIterationArgs <T> e)
{
double dfPct = (double)e.Iteration / m_nMaxIteration;
m_progress.SetProgress(dfPct);
}
