how to optimize neural network thyroid example using genetic algorithm?

i'm trying to to optimizing thyroeid example by pattern recognition .

i wana modify hidden layer size , and i want it from ga function ,

in default way is : hiddenLayerSize = [4 5 ] ; net = patternnet(hiddenLayerSize);

now i want hiddenLayerSize get from ga func , that gaves me minimum and optimal number of neuron .

is there anywat to omplent this case, tnx and srry for my broken english (

Answer 1

[Greg Heath]

Are you sure you want the complexity of 2 hidden layers?

One is sufficient.

How many input/target pairs? N =

Input vector dimensionality? I =

Number of classes ? c =

Are target columns also columns of eye(c)?

Default datadivision? Ntrn/Nval/Ntst = (0.7/0.15/0.15)*N

Are the No. of training equations Ntrneq = Ntrn*c

greater than number of unknown weights?

 Nw = (I+1)*H+(H+1)*c      % One hidden layer

 Nw = (I+1)*H1+(H1+1)*H2+(H2+1)*c % Two hidden layers?

Search both the NEWSGROUP and ANSWERS using

 PATTERNNET GENETIC

and

 PATTERNNET GA

Hope this helps.

Thank you for formally accepting my answer

Greg

PS: Searching with GA or Genetic and NEURAL, FITNET or FEEDFORWARDNET might help also.

[koorosh]

thank's for your help ,

look , as you know thyroid example have thyroidInputs - a 21x7200 matrix consisting of 7200 patients characterized by 15 binary and 6 continuous patient attributes. thyroidTargets - a 3x7200 matrix of 7200 associated class vectors defining which of three classes each input is assigned to. Classes are represented by a 1 in row 1, 2 or 3 ;1. Normal, not hyperthyroid 2. Hyperfunction 3. Subnormal functioning

i changed my mind and set hiddenLayer to one , now my case is to choose best size for this hidden layer by help of ga . ( 2?3?4?5?....? )

I've tried and create function named function percentErrors=Kfitfun(hiddenLayerSize)

and statement in commond windows like ga(kitfun,1) , but it does'nt work .

[Greg Heath]

I am using a borrowed computer without MATLAB. I don't even know if I have the GA Toolbox on my own machine.

Anyway, I have done some limited NEURAL searches with GENETIC and GA. Somewhat dissapointing. However, the searchword EVOLUTIONARY has proved to be a little more fruitful.

More later.

Greg

[Greg Heath]

WHOOPS: I meant to say PATTERNNET GA & GENETIC SEARCHES IN ANSWERS.

Changing PATTERNNET to NEURAL makes a HUGE difference.

SORRY

GREG