MATLAB: Patternnet for classification to 5bit targets

5bit codingDeep Learning Toolboxpatternnet

I have 30 classifications in my target.

Everything that I've searched regarding patternnet uses 1-of-C or 1-of(C-1) representations only.

I MUST use 5 output neurons only: I'm using 5 bit-representation for each category.

Can I do this using patternnet? If yes, how? If no, what is the best network to use here in Neural Network toolbox, and how? Will a simple feedforwardnet suffice?

Thanks in advance!

Best Answer

I remember Warren Sarle, the maintainer of the CANN FAQ lecturing and/or demonstrating the folly of using regular binary coding instead of 1-of(c or c-1). I thought it was in the FAQ. However, if you didn't find it, it must have been in a CANN post. Sorry for the missdirect.

As far as your training problem, it should be straightforward: just replace the column of eye(30) with the corresponding 5-dimensional 2 bit binary code.

You just need to find the functions that convert the elements of class = [1:32] to the corresponding columns of the 2-bit binary matrix with size(target) = [ 5 32 ]. ... and vice versa.

It sounds like your real problem stems from the post training algorithms of patternnet assuming the 1-of-c coding.

In that case you probably have to write your own or modify the source code of the MATLAB routines.

Modifying the nnet functions used to be very easy. Now it is rather complicated because of voluminous overhead. However, you could wade thru all of the overhead to find the important statements to modify. Then either keep the overhead or use a stripped down version.

Also, check the source code of patternnet to see what changes are made before and after it's call to feedforwardnet. You may just want to use feedforwardnet with defaults overwritten.

Hope this helps.

Greg

Related Solutions

MATLAB: How to get better test error/accuracy with neural networks

 1. Is there evidence that Hopt could be more than 10? i = 1:2:19 
2. Since your data set is huge, why not use tic and toc to time your runs
3. Why are you complicating the code by specifying net properties and values that are already defaults?
4. Your comments say TRAINSCG, the default for patternnet, and recommended for binary outputs but your code uses TRAINRP. Are you having size problems with TRAINSCG?
5. Your Plot options are those for regression, not classification.
    % see those associated with patternnet ;
    net = patternnet % NO SEMICOLON
6. Your standarizations are incorrect. Use ZSCORE or MAPSTD.
    Check to make sure EACH variable is standardized
7. Unfortunately, I'm not familiar with PLS (although it is the correct 
function to use for classifier input variable reduction) so, some of the 
following advice may be questionable
8. Are you trying to reduce 64 dimensions to 8?
9. XL and YL should be transposed
10. You could save the weights using getwb instead of or in addition to 
saving the nets.
11. Save and plot the overall and trn/val/tst/ performances vs numhidden
12. Modify to calculate , save and plot, overall and trn/val/tst/ percent 
classification errors.
13. To mitigate the probability of poor initial weights, consider a double loop design where the inner loop is over Ntrials different weight intializations for each value of numhidden. I use this technique almost all of the time. Search in NEWSGROUP and ANSWERS for examples using
 greg patternnet Ntrials

Hope this helps.

Thank you for formally accepting my answer

Greg

MATLAB: How to do testing and training using feed forward neural network

Use patternnet, not feedforwardnet. The target matrix should be a row vector containg zeros and ones indicating the true class of the corresponding input.

help patternnet

doc patternnet

Hope this helps.

Greg

Best Answer

Related Solutions

MATLAB: How to get better test error/accuracy with neural networks

MATLAB: How to do testing and training using feed forward neural network

Related Question