net.predictFcn produces 'Argument to dynamic structure reference must evaluate to a valid field name.'

The following code works great, i.e. I get the correct result an d a good performance score. How ever when I use the net.predictFcn on a new matrix or in fact on the trining data I get the 'Argument to dynamic structure reference must evaluate to a valid field name.' error. I've looked at other example from people who have got the same error but cannot work out the issue. The code snippit is :-

Inps = readtable("D:\RacingFormBook\Res.xlsx", opts, "UseExcel", false);


%% Clear temporary variables
clear opts

Outs=Inps(:,11);
Inps = removevars(Inps, 'FPos');

OutsM=table2array(Outs);
OutsM=OutsM';

%Convert catergorical to numeric next using unique

[GN, ~, CourseC] = unique(Inps(:,3));
[GN, ~, GoingC] = unique(Inps(:,4));
[GN, ~, HorseC] = unique(Inps(:,5));
[GN, ~, JockeyC] = unique(Inps(:,6));
[GN, ~, TrainerC] = unique(Inps(:,9));

Inps = removevars(Inps, {'Course','Going','HorseName','Jockey','Trainer'});
InpsM = table2array(Inps);
InpsMC=cat(2,InpsM,CourseC,GoingC,HorseC,JockeyC,TrainerC);
InpsMC=InpsMC';

clear GoingC HorseC TrainerC CourseC JockeyC GN Inps Outs InpsM

InpsMCG=gpuArray(InpsMC);
OutsMG=gpuArray(OutsM);

net=feedforwardnet([60,30,1]);
net.trainFcn = 'trainscg';
net=train(net,InpsMCG,OutsMG);

Then I try the net.predictFcn on InpsMCG or another set of imported results and get the error. I'm really stumped so has anyone got any ideas?

SPG

Answer 1

[Stephen Gray]

Hi all. Solved it myself. I should have been using net(matrix) instead of predictFcn to get the results. Error is a bit misleading though.

SPG