How do you fill in NaN in time series using neural networks?

I am having difficulty filling in NaN values with time series data with neural networks. The simulated program is below, created using the Neural Network time series wizard. The following error appears when using yg=netc(num2cell(biomet,2)); or yg=net(num2cell(biomet,2)).

Error using network/sim (line 271) Number of inputs does not match net.numInputs.

Error in network/subsref (line 16) otherwise, v = sim(vin,subs{:});

   N = 700 ;     % Number of data points
   biomet= 0:0.5:350;
   biomet2=0:0.25:175;     % Generate a random data
   biomet=[cos(biomet(1:700));sin(biomet2(1:700))]';
    frac_gaps = 0.3 ;   % Fraction of gaps 
   gaps = randsample(1:N,frac_gaps*N) ;  % gaps position
   target=sin(1:700)';
   target(gaps) = NaN ;     % Replace gaps with NaN
   time=1:700;
   time=1:700;
  %
  % Solve an Autoregression Problem with External Input with a NARX Neural Network
  % Script generated by Neural Time Series app
  % Created 15-Apr-2017 16:27:08
  %
  % This script assumes these variables are defined:
  %
  %   biomet - input time series.
  %   target - feedback time series.
  %
  X = tonndata(biomet,false,false);
  T = tonndata(target,false,false);
  %
  trainFcn = 'trainlm';  % Levenberg-Marquardt backpropagation.
  %
  % Create a Nonlinear Autoregressive Network with External Input
  inputDelays = 1:2;
  feedbackDelays = 1:2;
  hiddenLayerSize = 10;
  net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize,'open',trainFcn);
  %
  % Prepare the Data for Training and Simulation
  [x,xi,ai,t] = preparets(net,X,{},T);
  %
  % Setup Division of Data for Training, Validation, Testing
  net.divideParam.trainRatio = 60/100;
  net.divideParam.valRatio = 20/100;
  net.divideParam.testRatio = 20/100;
  %
  % Train the Network
  [net,tr] = train(net,x,t,xi,ai);
  %
  % Test the Network
  y = net(x,xi,ai);
  e = gsubtract(t,y);
  performance = perform(net,t,y);
  %
  % View the Network
  %view(net)
  %
  % Closed Loop Network
  netc = closeloop(net);
  netc.name = [net.name ' - Closed Loop'];
  %view(netc)
  [xc,xic,aic,tc] = preparets(netc,X,{},T);
  yc = netc(xc,xic,aic);
  closedLoopPerformance = perform(net,tc,yc);
  %
  % Step-Ahead Prediction Network
   nets = removedelay(net);
   nets.name = [net.name ' - Predict One Step Ahead'];
  % view(nets)
   [xs,xis,ais,ts] = preparets(nets,X,{},T);
   ys = nets(xs,xis,ais);
   stepAheadPerformance = perform(nets,ts,ys)
  %
  yg=net(num2cell(biomet,2));
  %
  figure(1)
  clf
  scatter(target,y)
  %
  figure(2)
  clf
  plot(time,target)
  hold on
  plot(time,y)

Thanks.

Answer 1

[Greg Heath]

Try again using real data. Using random data for a time series makes no sense at all: It is the correlation between adjoining points that allows timeseries to be useful.

In addition to the examples used in the documentation

 help narxet

 doc narxnet

there are several more data sets obtainable using the command

 help nndatasets.

Also, I have posted many examples in both the NEWSGROUP and ANSWERS. Just search with

 greg narxnet

Hope this helps.

Thank you for formally accepting my answer

Greg

[bikefan]

Thanks, I changed the input to be actual data. However, when I try to gap fill the data (also by checking your many responses), and the neural networks preformed more expectantly.

However, I tried using y=net(x), and still had NaN in the gap filled dataset. The original 'x data' had no gaps or NaN values.

Also the resultant matrix was 2 entries shorter. How would I complete a gap fill with multiple inputs and preserve the original matrix dimensions?