Deploying standalone executable on Apache

I've done installing apache, setting up, tested with few algorithm given by mathworks, like mycgimagic.m. It works perfectly.

Now, I would like to deploy this algorithm, a neural network classifier to the apache web server, anyone can help with me what line i should add so that I can pass the answer correctly to the end users?

The neural network classifier,

%taking input
fn=input('Enter WAV filename : ','s');
[wave,fs]=wavread(fn);
newsampled= resample(wave,44100,fs); %resampling
t=0:1/44100:(length(newsampled)-1)/44100;

%windowing
windowed = newsampled(1:50000); 

wd=windowed;
% Do a multi-level analysis to ten levels with the
% Daubechies-4 wavelet
[C,L] = wavedec(wd,10,'db4');  
A1 = wrcoef('d',C,L,'db4',1);   % Reconstruct the details at various levels
A2 = wrcoef('d',C,L,'db4',2);
A3 = wrcoef('d',C,L,'db4',3);
A4 = wrcoef('d',C,L,'db4',4);
A5 = wrcoef('d',C,L,'db4',5);
A6 = wrcoef('d',C,L,'db4',6);
A7 = wrcoef('d',C,L,'db4',7);
A8 = wrcoef('d',C,L,'db4',8);
A9 = wrcoef('d',C,L,'db4',9);
A10 = wrcoef('d',C,L,'db4',10);

%Feature Extraction by using sum of absolute divided by total sample
feat1 = sum(abs(wd))/50000
feat2 = sum(abs(A1))/50000
feat3 = sum(abs(A2))/50000
feat4 = sum(abs(A3))/50000
feat5 = sum(abs(A4))/50000
feat6 = sum(abs(A5))/50000
feat7 = sum(abs(A6))/50000
feat8 = sum(abs(A7))/50000
feat9 = sum(abs(A8))/50000
feat10 = sum(abs(A9))/50000
feat11 = sum(abs(A10))/50000

featset1=[feat1 feat2 feat3 feat4 feat5 feat6 feat7 feat8 feat9 feat10 feat11]

%train Neural Network
P=[ 0.0756  0.0172  0.0283  0.0638 
  0.0000  0.0000  0.0000  0.0000 
  0.0002  0.0002  0.0001  0.0001  
  0.0005  0.0008  0.0004  0.0005  
  0.0004  0.0012  0.0008  0.0007  
  0.0007  0.0013  0.0049  0.0023  
  0.0021  0.0025  0.0153  0.0105 
  0.0071  0.0075  0.0132  0.0363
  0.0222  0.0123  0.0078  0.0454  
  0.0434  0.0060  0.0045  0.0109 
  0.0470  0.0011  0.0009  0.0027  ] %training feature set

T=eye(4)%identity matrix of 4x4

net = newff(P,T,20,{'tansig','logsig'});
net.performFcn = 'mse';%mean square error
net.trainParam.epochs = 100;
net.trainParam.mc = 0.98;%momentum coefficient
net.trainParam.lr=0.005;%Learning rate
net.trainParam.lr_inc=1.06; %Ratio to increase learning rate
net.trainParam.lr_dec=0.7; %Ratio to decrease learning rate

%train the network
net = train(net,P,T);
y = sim(net,P); %simulate again with same input

%setting featset1 as input to the neural network
Po=transpose(featset1);
newnet = net;
res1 = sim(newnet,Po);
disp('res1:');
disp(res1);

%checking the index of class
[Co,I] = max(res1);

%data comparison
if (I == 1)
  disp('This is a signal of Normal Heart Sound')
elseif (I == 2)
  disp('This signal is classified as illness: Mitral Stenosis')
elseif (I == 3)
  disp('This signal is classified as illness: Aortic Regurgitation')
elseif (I == 4)
  disp('This signal is classified as illness: Ventricle Septic Defect')
else
  disp('Error: No Classification is available')
end

-My understanding is I have to add some lines for generating the header and footers before compiling with "mcc" so that it can pass the answer and show it on the web browser.

-I'm still new with programming with C/html, anyone can help me on this task? I'm in dire need of help and would very appreciate if someone can enlighten me on this.

Answer 1

[Chirag Gupta]

One thing to note:

I don't think you can compile (with mcc) the neural network train function. You can only use pre trained nets with compiled application.

http://www.mathworks.com/products/compiler/compiler_support.html

Answer 2

[Justin Yau]

okie, so i'll be using a save function to save a pre-train neural network and deploy it using a load function. Thx for the heads up!~

But in this case, will the standalone executable be able to communicate with the neural network made? as in the neural network have to save separately as another .mat file rite?

Anyway, I still need some guidance on making lines for return answers in web browser, from the example i saw give by mathworks, it was embedded with some html coding to print header, body and footer for passing back result via web browser.

Answer 3

[Chirag Gupta]

Yes. The standalone executable will be able to communicate with the MAT file. You should compile it:

mcc -m myfunc.m -a mynetwork.mat

or just add the MAT file as an additional file in the deploytool project.

In your code, you should be able to load the file using the load command.

I don't have much idea about executing a standalone EXE from APACHE, however this can also be done using MATLAB Builder JA product (example is here):

http://www.mathworks.com/help/toolbox/javabuilder/ug/bsqrikz-26.html

[Justin Yau]

thx for the reply again, thx for teaching that line of mcc =)

by deploying standalone EXE from APACHE, i just deploy the .exe and mcr into "cgi-bin" thus, create a simple HTML interface to put in "htdocs" and the user input will communicate with the .exe by using "FORM ACTION" line. That's what i know about that.

Is any programming skill required for MATLAB Builder JA? cause I have shallow knowledge about java, i'll look into it anyway.

I hope i can figure out what html line i need to add to print the header/body/footer and result as soon as possible, deadline is very near.

Anyway, I really appreciate your help =)! thx~