One step ahead prediction with Recursive Neural Net (RNN)

Hello, I am trying to use MATLABS RNN function layrecnet to do one step ahead prediction. But it does not let me use "removedelay" as in other examples, resulting in an error.

There must be another way to do it, I am wondering how to get one-step ahead prediction working with the RNN?

    neto = layrecnet(inputDelays,hiddenLayerSize);
    [Xo,Xoi,Aoi,To] = preparets(neto,Xorig,Torig);
    [ neto, tro, Yo, Eo, Xof, Aof ] = train(neto,Xo,To,Xoi,Aoi);
    view(neto)
    Yo = neto(Xo,Xoi,Aoi);

    to = cell2mat(To);
    MSE00o = mean(var(to',1)) % Normalization Referenc
    NMSEo = mse(Eo)/MSE00o
    R2o   = 1 - NMSEo           
    yo = cell2mat(Yo);

    nets = removedelay(neto);

>> nets = removedelay(neto); Error using removedelay (line 57) Removing 1 to input delays would result in a negative input weight delay.

The layrecnet help mentioned using removedelay to do prediction, but I think the help is wrong.

>> help layrecnet layrecnet Layered recurrent neural network.

   Layer recurrent networks with two (or more) layers can learn to
   predict any dynamic output from past inputs given enough hidden
   neurons and enough recurrent layer delays.

   layrecnet(layerDelays,hiddenSizes,trainFcn) takes a row vectors
   of layers delays, a row vector of hidden layer sizes, and a
   backpropagation training function, and returns a layer recurrent neural
   network with N+1 layers.

   Input, output and output layers sizes are set to 0.  These sizes will
   automatically be configured to match particular data by train. Or the
   user can manually configure inputs and outputs with configure.

   Defaults are used if layrecnet is called with fewer arguments.
   The default arguments are (1:2,10,'trainlm').

   Here a layer recurrent network is used to solve a time series problem.

     [X,T] = simpleseries_dataset;
     net = layrecnet(1:2,10);
     [Xs,Xi,Ai,Ts] = preparets(net,X,T);
     net = train(net,Xs,Ts,Xi,Ai);
     view(net);
     Y = net(Xs,Xi,Ai);
     perf = perform(net,Y,Ts)

   To predict the next output a step ahead of when it will occur:

     net = removedelay(net);
     [Xs,Xi,Ai,Ts] = preparets(net,X,T);
     Y = net(Xs,Xi,Ai);

   See also narxnet, timedelaynet, distdelaynet.

    Reference page in Help browser
       doc layrecnet

Answer 1

[Greg Heath]

 The removedelay command reduces all delays by 1

 However feedback delays must be positive

 Therefore, removedelay will cause an error if the minimum feedback 
 delay of any net is 1.

 This includes nar and narx. 

 Similarly, input delays must be nonnegative 

 Therefore, removedelay will cause an error if the minimum input 
 delay of any net is 0.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Alexandra Sikinioti-Lock]

So does this mean that RNNs (layrecnet) cannot be used for one step ahead predictions? The reason I am asking again is because the syntax for an ANN is layrecnet(layerDelays,hiddenSizes,trainFcn) and the input delay cannot be inserted, from what I understand it is always zero. Could you perhaps provide an alternative for one step ahead prediction with RNN, if it exists?

[Greg Heath]

 See the documentation:

   help narxnet
   doc narxnet

 net = narxnet(ID,FD,H);

ID >=0, FD>=1, H = [] (for H = 0), or H >=1

 DEFAULT VALUES

 net = narxnet;

% ID = FD = 1:2 and H = 10

Hope this helps.

Greg