Question

variable input is a jagged array, i.e. 90 x 600.

ideal is also jagged array of 90 x 1

BeginTraining()
{
    FeedforwardNetwork network = new FeedforwardNetwork();
    network.AddLayer(new FeedforwardLayer(600));
    network.AddLayer(new FeedforwardLayer(1000));
    network.AddLayer(new FeedforwardLayer(90));
    network.Reset();

    // train the neural network
    Train train = new  HeatonResearchNeural.Feedforward.Train.Backpropagation.Backpropagation(network, input, ideal,
                0.7, 0.9);

    int epoch = 1;

    do
    {
        lblNoEpochs.Text= epoch.ToString();
        lblNoEpochs.Update();
        train.Iteration();
        lblPcError.Text = train.Error.ToString();
        lblPcError.Update();
        epoch++;
    } while((epoch < 5000) && (train.Error > 0.001));

        // test the neural network
        for (int i = 0; i < input.Length; i++)
        {
            //testing code
        }

Each epoch s taking at least 5 minutes.

How to reduce this time?

Answer 1

I think the real question is – can the do/while be parallel.

as long as I know the training waves depends on each other so the next depends on the results of the previous. hence ordering is absolutely a must.

so I figure the answer is – this cannot be faster.