How to determine the neuron in LVQ

Dear all

I have one question, now i'm using LVQ to classify fruits but i confuse to determine the neuron that I must use. I have matrix, rows = 15 and the column = 60, the class = 3. How many neuron I must use.

Thank you

Answer 1

[Greg Heath]

[ I N ] = size(input) % [15 60]

[ O N ] = size(target) % [ 3 60 ]

For lvqnet net.divideFcn = ''. Therefore,

 Ntrn = N               %60 Training examples

 Ntrneq = N*O           %180 Training equations

For a single hidden layer with H units

 Nw = (I+O)*H            % No. of unknown weights to estimate

For a robust design desire Ntrneq >> Nw. or H << Hub where

 Hub  -1 + ceil( Ntrneq/(I+O))  % 9

Since H << Hub is not possible, try reducing I and/or increasing N (e.g., adding noisy copies of the originals and making H as small as possible.

You may also wish to investigate 10-fold cross-validation and/or another type of classifier (e.g., patternnet or newrb).

Hope this helps.

Thank you for formally accepting my answer

Greg

[Doni]

Thank you for your answer, but i' still confuse with it. Is there any equations to determine the number of neuron that we use? Thank you