How do I find out the number of neurons in layers?

Dear all,

I have this code for training neural network (RBF) :

load('trenovaci_modely1_velky')
disp('Trénovací modely byly načteny.')
P = [velky_tvar{1,:}];
T = [velky_tvar{2,:}];
net = newrb(P,T,0,0.3)
save net 
disp('Neuronová síť byla uložena.')

I need to know how many neurons are in layers? Does anyone have any idea? Thank you for your answers.

Answer 1

[Greg Heath]

% newrb( x, t ,MSEgoal, spread, Nbmax, dNdisp )

% x - I x N matrix of N "I"nput vectors.

% t - O x N matrix of N "O"utput target vectors.

% MSEgoal - Mean squared error goal, default = 0.0.

% spread - Spread of radial basis functions, default = 1.0.

% Nbmax - Maximum number of basis neurons, default is N.

% dNdisp - Number of neurons to add between displays, default = 25.

'BUG: Output listing WILL SKIP line for neurons = 1 '

 close all, clear all, clc
 [ x, t ] = simplefit_dataset;
 [ I  N ] = size(x)  % [ 1 94 ]
 [ O  N ] = size(t)  % [ 1 94 ]

 zx = zscore( x',1 )'; % Standardize to zero-mean/unit-variance
 zt = zscore( t',1 )';
 figure(1)
 plot( zx, zt )
 hold on

 MSEgoal = 0.01*mean(var(zt',1)) % 0.01
 spread  = 1
 Nbmax   = N - 2*round(0.15*N)   % 66 ~ 0.7*N
 dNdisp  = 1

 [ net tr ] = newrb( zx, zt, MSEgoal, spread, Nbmax, dNdisp);

 % NEWRB, neurons = 0, MSE = 1
'    'BUG: SKIPS neurons =  1 OUTPUT LISTING '
% NEWRB, neurons = 2, MSE = 0.272804
% NEWRB, neurons = 3, MSE = 0.267621
% NEWRB, neurons = 4, MSE = 0.181115
% NEWRB, neurons = 5, MSE = 0.0811973
% NEWRB, neurons = 6, MSE = 0.0246589
% NEWRB, neurons = 7, MSE = 0.0115911
% NEWRB, neurons = 8, MSE = 0.00415462

 epochs = tr.epoch;  MSE= tr.perf;
result   = [ epochs' MSE' ]
% result = epoch     MSE
%            0            1
%            1      0.29336
%            2       0.2728
%            3      0.26762
%            4      0.18112
%            5     0.081197
%            6     0.024659
%            7     0.011591
%            8    0.0041546

 y = net(zx);
 figure(1)
 hold on
 plot( zx, y, 'r.')
 e    = zt-y;
 NMSE = mse(e)  %0.004154

 Nw = net.numWeightElements  % 25

 H  = (Nw-O)/(I+1+O)         % 8
 H = size(net.IW{1},1)
 H = size(net.b{1},1)
 H = size(net.LW{2},2)

Hope this helps

Thank you for formally accepting my answer.

Greg

Answer 2

[Vinod Sudheesh]

You could do this by querying the "size" property of each of the individual neural network layers. For example, please see the code snippet below:

>> net=feedforwardnet([10 11 12]);

>> net.layers{1}.size

>> net.layers{2}.size

>> net.layers{3}.size

[Greg Heath]

Not for NEWRB!

If [ I N] = size(input) [ O N ] = size(target)

the initial pretraining topology is I-O and the final postraining topology will be I-H-O.

However, H is unknown until the training is complete.