MATLAB: Can somebody explain me how to use “divideind”?

Deep Learning Toolboxdivideindneural network

I am using Neural Network tool box for pattern recognition. The tool box uses random values of the input matrix for training test validation of defined percentage which results in different performance graph every time i train it.
I read that if i generate the advanced script and use divideind i can fix the matrix of validation,testing and training. But i'm not sure how to use it and what amendments should be made in the advance script. Kindly Help.
P.S dont tell me to read help and doc its use less (atleast for me)

Best Answer

The first time use as many defaults as possible. Defaults and basic code examples are listed in
 help patternnet and 
 doc patternnet
Also, if you remove the semicolon, all the defaults will be revealed via
 net = patternnet(hiddenLayerSize)
Once you are error free, start to make changes.
Even the correct code may not work because of an unfortunate set of initial weights. Therefore, with the correct number of hidden nodes I usually design 10 nets in a loop. The best is chosen by the lowest validation set error. The prediction of performance on unseen data (generalization) is obtained from the corresponding test set error.
Often the default number of hidden nodes(10) is nonoptimal. My solution is a double loop design with the outer loop over ~ 10 candidate values.
Details are in my posts obtained from searching
 greg patternnet Ntrials
Hope this helps.
Thank you for formally accepting my answer
Greg
P.S. More later.
Related Question