About neural network : newbie question

I am a new to matlab. I just covert my image to matrix and want to train it. here is my coding:

P= [];
for i =1:15

str=strcat('E:Skripsi\crop2\image',int2str(i),'.bmp');
eval('img=imread(str);');
 img=reshape(img,[],1);
    P= [P img];
end
T=[1 1 1 1 1 1 1 1 1 1 0 0 0 0 0];// if true image i make it 1 , negative image make it  0
p = num2cell(P,1);
t = num2cell(T,1);

net = newff(minmax(p), [3 1], {'logsig' 'purelin'});
net.trainParam.epochs = 10;
net = train(net,p,t);
Y=sim(net,p)

these is what i got when i compile it

 [0.9794]    [0.9974]    [1.2784]    [1.0014]    [1.0058]    [0.9904]    [0.7293]    [1.0001]    [1.0106]

Columns 10 through 15

    [1.0002]    [0.0618]    [3.3131e-04]    [9.9559e-04]    [0.0134]    [0.0178]

then my problem is what should i do to make the result success i train in percentage with error below 1% (what i want: example i train 10 true image to 1 and one result of my train is 0.798 then accurate =90%) thanks for the help.

Answer 1

[Greg Heath]

There is no reason to use cells or only train for 10 epochs.

Use round to convert the output to {0,1}

Thank you for formally accepting my answer

Greg

[Parkz Leingodz]

firstly thanks for your guidence, i'm still don't understand with "no reason to use cells " so what should i do for best? do you have better solution? because i tried once without cell and it wents error "inputs are not a matrix or cell array with a single matrix" thank you for kindly help

[Greg Heath]

No reason to use cells: Just use P and T instead of p and t.

You have successfully gotten a cell answer with non-integer values. However, to get {0,1} you have to use cell2mat and then round. So why use cells in the first place?

I cannot answer your last question because I haven't see your noncell code. What dimension is P?

[Parkz Leingodz]

here is my non cellcode at first :

P= [];
for i =1:15

str=strcat('E:Skripsi\crop2\image',int2str(i),'.bmp');
eval('img=imread(str);');
 img=reshape(img,[],1);
    P= [P img];% convert all the image to one matrix
end
T=[1 1 1 1 1 1 1 1 1 1 0 0 0 0 0]; 

net = newff(minmax(P), [3 1], {'logsig' 'purelin'});
net.trainParam.epochs = 10;
net = train(net,P,T);
Y=sim(net,P)

and if i compile it , here is the error :

Error using newff>new_5p0 (line 257)
Inputs are not a matrix or cell array with a single matrix.

Error in newff>create_network (line 128)
  net = new_5p0(varargin{:});

Error in newff (line 102)
  out1 = create_network(varargin{:});

Error in f (line 12)
net = newff(minmax(P), [3 1], {'logsig' 'purelin'});

actually i don't know how to fix this error and where's the error, so i just try to change P to cell and it works.

but what u mention earlier is true, i have to change cell2mat when i want get {0,1}. so do you have an advice to fix my non cell code?

i don't know how to answer what dimension P is. i just want to convert image to one matrix and train it. thanks

[Greg Heath]

whos P

size(P)

[Parkz Leingodz]

size P is 16x21 pixels.

P is a matrix that save whole matrix value of img.