Which Network can I use for cancer diagnosis?

Hello,I am new in MATLAB application and I have some doubts and queries also.Please suggest me.

1> I am using NN toolbox nprtool for cancer diagnosis.I want to do some comparative test using other networks and algorithms.Which are the all networks and algorithms can I use, please suggest and also which tool can I use?

2>nntool also have some networks but I am getting better result using nprtool for BPNN, why so?

3> Nprtool have the option for BPNN,is it possible to use other networks also like wise?if so,then what is the command for that?

4> Does the code for one release of MATLAB varies from other?I am using R2008a.Is it possible se the same come for all releases?

Thanks a lot. Shraboni

Answer 1

[Greg Heath]

Which Network can I use for cancer diagnosis? Asked by Shraboni Mondal on 29 Dec 2012 at 15:30

% Hello,I am new in MATLAB application and I have some doubts and queries
% also.Please suggest me.
% 
% 1> I am using NN toolbox nprtool for cancer diagnosis.I want to do some
% comparative test using other networks and algorithms.Which are the all
% networks and algorithms can I use, please suggest and also which tool can
% I use?

Tool: nprtool

Nets: Either patternnet or the obsolete newpr.

help newpr

doc newpr

help patternnet

doc patternnet

% 2>nntool also have some networks but I am getting better result using % nprtool for BPNN, why so?

The 'pr' indicates created for pattern recognition and classification

% 3> Nprtool have the option for BPNN,is it possible to use other networks % also like wise?if so,then what is the command for that?

Forget it. Only use patternnet or newpr

% 4> Does the code for one release of MATLAB varies from other?

Yes

% I am using R2008a. Is it possible se the same come for all releases?

No. Check the website documentation for BUGS and release changes

Patternnet introduced in 2010. Use newpr. Practice on the 7 sample data sets

InputMatrix: Standardize input matrix with zscore or mapstd

                   [ I N ]  =  size(input)

TargetMatrix: For c classes the target matrix columns should be columns of the c-dimensional unit matrix.

                   [ O  N ]  =  size( target)   % O = c

Classindices: classind = vec2ind(target)

Transfer Functions: { 'tansig' 'logsig' }

Training Function: 'trainscg'

Training: [ net tr ] = ntrain(net,inputs,target);

History: tr = tr

Output: y = net(input);

Assigned classes: assignedclass = vec2ind(y)

Errors: Err = (assignedclass ~= classind)

No. errors: Nerr = sum(Err)

Error Rate: PCTerr = 100*Nerr/N

To obtain separate trn/val/tst errors, use the masks and indices in tr.

Hope this helps.

Thank you for formally accepting my answer.

Greg

[Shraboni Mondal]

Thanks a lot Mr. Greg Health. your suggestions are great !

[Greg Heath]

Glad you liked them.

Dr. Greg Heath