How can i train neural network for letter recognition using matlab and she to has 26 outputs like 26 letters in english alphabet?

Hi all. I want to make project for letter recognition data using neural network. I use this dateset: https://archive.ics.uci.edu/ml/datasets/Letter+Recognition Тhis is my code to prepare data:

        clear all;
        fid = fopen('letter-recognition.data', 'rt');
        num_attrib = 16;
        fmt = ['%s', repmat('%f', 1, num_attrib)];
        datacell = textscan(fid, fmt, 'Delimiter', ',', 'CollectOutput', 1);
        fclose(fid); 
        which_letter = char(datacell{1}); 
        attribs = datacell{2}; 
        target_codes = which_letter - 'A' + 1;
        train_set = attribs(1:end-4000, :); 
        train_targets = target_codes(1:end-4000);

Тhe problem that I have is: I can not do to have 26 targets (as there are letters in the alphabet). I wanna to have 16000 inputs and 26 outputs.

Answer 1

[Greg Heath]

The target columns should be columns of the unit matrix

         eye(26)

The relationships between the target and class indices are given by

   target = ind2vec(classindices)

   classindices = vec2ind(target)

The output yields the estimated indices

   estimatedindices = vec2ind(output)

 Nerrs = numel(estimatedindices~=classindices)

Hope this helps.

Thank you for formally accepting my answer

Greg

[Ady]

Hi Greg. Thanks for the answer. Can you show me how i can do this with example involving current my code, because i tried but i couldn't do it.

[Greg Heath]

I don't understand your code. Cut and paste the following:

% classindices

 clind = [ 5 3 1 4 2]
 N     = length(clind) 
target = full(ind2vec(clind))
...
output = target+0.5*randn(5,5)

% estimated indices

 estind = vec2ind(output)
 Nerr   = sum(estind~=clind)
 PctErr = 100*Nerr/N

%Hope this helps

%Greg