Query on Neural network results
1 view (last 30 days)
Show older comments
I am working on handwritten character recognition. I have trained the neural network with (130 * 85) inputs and (130 *26) targets. I am using the nprtool in matlab and here are the results which I am getting with 50 hidden neurons
Training : 0 % error validation : 0% error Testing : 5 % error
But, I am getting these results after lot of retraining. So will I be able to classify the characters with these results?
0 Comments
Accepted Answer
Greg Heath
on 19 Apr 2014
I = 85, O = 26, N = 130, H = 50
[ I N ] = size(input) % [ 85 130]
[ O N ] = size(target) % [ 26 130]
Ntrn = N -2*(0.15*N) % 91 training examples
Ntrneq= Ntrn*O % 2366 training equations
Nw = (I+1)*H+(H+1)*O % 5626 unknown weights
Hub = -1 + ceil( (Ntrneq-O) / ( I + O + 1 ) % 20
Try to lower H as much as possible. It should make your model more robust (w.r.t. noise, measurement error and unseen data). Typically, I try to look at ~10 different values for H and ~Ntrials = 10 different designs for each value of H to mitigate the probability of getting a poor set of initial weights.
I have many posts in the NEWSGROUP and ANSWERS. Search on
greg patternnet Ntrials
greg newpr Ntrials
for h = Hmin:dH:Hmax
....
for i = 1:Ntrials
....
end
end
Hope this helps,
Greg
0 Comments
More Answers (0)
See Also
Categories
Find more on Sequence and Numeric Feature Data Workflows in Help Center and File Exchange
Products
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!