Neural network input mistmatch (Iris Dataset) - matlab

Question

I currently have an error that i can't pass this is the short code and everything needed in order to have a general idea about my problem
clear;
close all; clear ;
load fisheriris;
m = meas;
d = num2cell(m);
d(:,5) = species(:,1);
c = cvpartition(d(:,5),'kfold',10);
CeDam = cell(10,1);
CeVrem = cell(10,1);
for i=1:10
CeDam{i} = [d(test(c,i),1) d(test(c,i),2) d(test(c,i),3) d(test(c,i),4)]';
end
for i=1:10
CeVrem{i} = d(test(c,i),5)';
end
for i = 1:10
a = CeVrem{i};
[n,m] = size(a);
for j = 1:n
for k = 1:m
if isequal(a(j,k),'setosa') a{n,m} = [1 0 0];
elseif isequal(a(j,k),'versicolor') a{n,m} = [0 1 0];
else a{j,k} = [0,0,1];
end
end
end
CeVrem{i} = a;
end
net = newff(cell2mat(minmax(CeDam{1})),[3 3 3],{'logsig','logsig','logsig',},'trainlm');
net.LW{2,1} = net.LW{2,1}*0.5;
net.b{2} = net.b{2}*2;
net.performFcn = 'mse';
net.trainParam.epochs = 100;
err = 0;
i = 1;
j = 1;
while i <= 10
while j <= 10
if i~=j net = train(net,CeDam{j},CeVrem{j});
end
j=j+1;
end
end
in the train part of the algorithm it gives me an input mistmatch which is very odd for me.
The error messages:
Error using trainlm (line 109) Number of inputs does not match
net.numInputs.
Error in network/train (line 106) [net,tr] =
feval(net.trainFcn,net,X,T,Xi,Ai,EW,net.trainParam);
i managed to fix everything after much work here is the code that works for anyone having the same problem in the future gl :D :).
clear;
close all; clear ;
load fisheriris;
m = meas;
d = num2cell(m);
d(:,5) = species(:,1);
c = cvpartition(d(:,5),'kfold',10);
CeDam = cell(10,1);
CeVrem = cell(10,1);
for i=1:10
CeDam{i} = [m(test(c,i),1) m(test(c,i),2) m(test(c,i),3) m(test(c,i),4)]';
end
for i=1:10
CeVrem{i} = d(test(c,i),5);
end
for i = 1:10
a = CeVrem{i}';
[n,m] = size(a);
b = zeros(3,m);
for j = 1:n
for k = 1:m
if isequal(a(j,k),{'setosa'}) b(1,k) = 1; b(2,k) = 0; b(3,k) = 0;
elseif isequal(a(j,k),{'versicolor'}) b(1,k) = 0; b(2,k) = 1; b(3,k) = 0;
else b(1,k) = 0; b(2,k) = 0; b(3,k) = 1;
end
end
end
CC{i} = b;
end
CC = CC';
net = newff(minmax(CeDam{1}),[3 3 3],{'logsig','logsig','logsig'},'trainlm');
net.LW{2,1} = net.LW{2,1}*0.6;
net.b{2} = net.b{2}*2;
net.performFcn = 'mse';
net.trainParam.epochs = 100;
errglob = 0;
i = 1;
j = 1;
while i <= 10
while j <= 10
if i~=j net = train(net,CeDam{j},CC{j});
end
j=j+1;
end
y=sim(net,CeDam{i});
y=round(y);
e = y - CC{i};
errcur=mse(net,CC{i},y);
errglob = errglob + mse(net,CC{i},y);
fprintf('Avem o eroare de %.2f pe foldul %d \n',errcur,i)
i=i+1;
end
errglob/10
this thread can be closed thx :)

Answer 1

I think you got some problems with mixing up cell and array formats...
Try to replace:
net = train(net,CeDam{j},CeVrem{j});
by:
net = train(net,cell2mat(CeDam{j}),cell2mat(CeVrem{j}')');
AND: please remove your infinite loops in i, by adding i=i+1; or replace the while loops by more natural for loops, e.g.
for i = 1:10
for j = 1:10
if i~=j
net = train(net,cell2mat(CeDam{j}),cell2mat(CeVrem{j}')');
end
end
end
AND: Where are you using your i inside the loop? I guess something is missing there...