Matlab NARX while loop

10 次查看（过去 30 天）
Tu Yuan Neoh
Tu Yuan Neoh 2016-8-26
回答： Greg Heath 2016-8-26
Hi,
I am running a MIMO NARX using limited data of 22 time steps for each I and O. To get around the inaccuracy of the prediction by the network due to data limitation, I plan to run a somewhat Monte Carlo-like experiment, where I retrain and reinitialize the training weights of the model for numerous times and plot the respective predictions to get a distribution. Each training set has to fulfill a set criteria of R and MSE values by using a while loop. I have my own doubts about this method as I am not an efficient coder. Initial attempts have some success but the while loop gets stuck frequently, i.e for the first four runs of the for loop I can get criteria fulfilling networks and the network exits the while loop but after that training while loop gets stuck, meaning it cannot exit the loop due to criteria non-conformance. I'm hoping someone can identify the flaws in my methodology and kindly correct them. Thanks!
for i=1:10
trainFcn = 'trainlm'; % Bayesian Regularization backpropagation.
% Create a Nonlinear Autoregressive Network with External Input
inputDelays = 1:2;
feedbackDelays = 1:2;
hiddenLayerSize=10;
net = narxnet(inputDelays,feedbackDelays,hiddenLayerSize,'open',trainFcn);
% Choose Input and Feedback Pre/Post-Processing Functions
% Settings for feedback input are automatically applied to feedback output
% For a list of all processing functions type: help nnprocess
% Customize input parameters at: net.inputs{i}.processParam
% Customize output parameters at: net.outputs{i}.processParam
net.inputs{1}.processFcns = {'removeconstantrows','mapminmax'};
net.inputs{2}.processFcns = {'removeconstantrows','mapminmax'};
% Prepare the Data for Training and Simulation
% The function PREPARETS prepares timeseries data for a particular network,
% shifting time by the minimum amount to fill input states and layer
% states. Using PREPARETS allows you to keep your original time series data
% unchanged, while easily customizing it for networks with differing
% numbers of delays, with open loop or closed loop feedback modes.
[x,xi,ai,t] = preparets(net,X,{},T);
% Setup Division of Data for Training, Validation, Testing
% For a list of all data division functions type: help nndivide
net.divideFcn = 'divideblock'; % Divide data randomly
net.divideMode = 'time'; % Divide up every sample
net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;
% Choose a Performance Function
% For a list of all performance functions type: help nnperformance
net.performFcn = 'mse'; % Mean Squared Error
% Choose Plot Functions
% For a list of all plot functions type: help nnplot
net.plotFcns = {'plotperform','plottrainstate', 'ploterrhist', ...
'plotregression', 'plotresponse', 'ploterrcorr', 'plotinerrcorr'};
% Test the Network
performance=1;% to setup an initial value for while loop
reg0=[0.5;0.5;0.5;0.5];% to setup an initial value for while loop
while (reg0(1)<0.9)|(reg0(2)<0.9)|(reg0(3)<0.9)|(reg0(4)<0.9)|(performance>0.01)
% 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);
[reg m b]=regression(t,y);
reg0=reg;
end
% Recalculate Training, Validation and Test Performance
trainTargets = gmultiply(t,tr.trainMask);
valTargets = gmultiply(t,tr.valMask);
testTargets = gmultiply(t,tr.testMask);
trainPerformance = perform(net,trainTargets,y)
valPerformance = perform(net,valTargets,y)
testPerformance = perform(net,testTargets,y)
% %%view the Network
%view(net)
% Plots
% Uncomment these lines to enable various plots.
%figure, plotperform(tr)
%figure, plottrainstate(tr)
%figure, ploterrhist(e)
%figure, plotregression(t,y)
%figure, plotresponse(t,y)
%figure, ploterrcorr(e)
%figure, plotinerrcorr(x,e)
% Closed Loop Network
% Use this network to do multi-step prediction.
% The function CLOSELOOP replaces the feedback input with a direct
% connection from the outout layer.
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)
% Multi-step Prediction
% Sometimes it is useful to simulate a network in open-loop form for as
% long as there is known output data, and then switch to closed-loop form
% to perform multistep prediction while providing only the external input.
% Here all but 5 timesteps of the input series and target series are used
% to simulate the network in open-loop form, taking advantage of the higher
% accuracy that providing the target series produces:
%%numTimesteps = size(x,2);
%%knownOutputTimesteps = 1:(numTimesteps-5);
%%predictOutputTimesteps = (numTimesteps-4):numTimesteps;
%%X1 = X(:,knownOutputTimesteps);
%%T1 = T(:,knownOutputTimesteps);
%%[x1,xio,aio] = preparets(net,X1,{},T1);
%%[y1,xfo,afo] = net(x1,xio,aio);
% Next the the network and its final states will be converted to
% closed-loop form to make five predictions with only the five inputs
% provided.
%%x2 = X(1,predictOutputTimesteps);
%%[netc,xic,aic] = closeloop(net,xfo,afo);
%%[y2,xfc,afc] = netc(x2,xic,aic);
%%multiStepPerformance = perform(net,T(1,predictOutputTimesteps),y2)
% Alternate predictions can be made for different values of x2, or further
% predictions can be made by continuing simulation with additional external
% inputs and the last closed-loop states xfc and afc.
% Step-Ahead Prediction Network
% For some applications it helps to get the prediction a timestep early.
% The original network returns predicted y(t+1) at the same time it is
% given y(t+1). For some applications such as decision making, it would
% help to have predicted y(t+1) once y(t) is available, but before the
% actual y(t+1) occurs. The network can be made to return its output a
% timestep early by removing one delay so that its minimal tap delay is now
% 0 instead of 1. The new network returns the same outputs as the original
% network, but outputs are shifted left one timestep.
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);
ys=transpose(cell2mat(ys));
;
for i=1:size(ys,2)
ypred(:,i)=(2.*ys(:,i)-1)*(minB(i)-maxB(i))+meanB(i);
end
%disp(ypred);
ypred0=[ypred0 ypred];% to store predictions
regstore=[regstore reg0];% to store performance
clearvars net
end

请先登录，再回答此问题。

采纳的回答

Greg Heath
Greg Heath 2016-8-26
The code you are using wastes time and space by making unneccesary assignments of variables to their default values. A better approach is to begin with the code that is in
help narxnet
doc narxnet
and
search the NEWSGROUP and ANSWERS for previously posted work.
For example
NEWSGROUP ANSWERS
narxnet 90 321
greg narxnet 71 251
A couple of quick look comments:
1. FOR loops are better than WHILE loops. Less chance of an infinite loop.
2. The loop is started in wrong place. All you have to loop over is number of hidden nodes and initial random weights. As a result most of your calculations are meaningless repetitions.
3. DIVIDEBLOCK is the correct datadivision mode for unbiased constant timestep prediction. Therfore your comment "Divide data randomly" is incorrect.
help divideblock
doc divideblock
4. You don't reveal any numerical information (I,O and resulting performance values)
5. A good openloop target for narxnet is R^2 >= 0.995 or R >= 0.9975. Your target of R>= 0.9 is woefully inadequate (R^2 > 0.81).
6. Take a look at some of my posts. Especially the NEWSGROUP posts over the past year or so.
Hope this helps.
Thank you for formally accepting my answer
Greg

更多回答（0 个）

类别

在 Help Center 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息

标签

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by