Drexel dragonThe Math ForumDonate to the Math Forum



Search All of the Math Forum:

Views expressed in these public forums are not endorsed by Drexel University or The Math Forum.


Math Forum » Discussions » Software » comp.soft-sys.matlab

Topic: Trainlm and Trainbr
Replies: 1   Last Post: Jul 28, 2013 7:02 AM

Advanced Search

Back to Topic List Back to Topic List Jump to Tree View Jump to Tree View   Messages: [ Previous | Next ]
Greg Heath

Posts: 5,955
Registered: 12/7/04
Re: Trainlm and Trainbr
Posted: Jul 28, 2013 7:02 AM
  Click to see the message monospaced in plain text Plain Text   Click to reply to this topic Reply

"Imran Babar" wrote in message <ksmmci$b6i$1@newscl01ah.mathworks.com>...
> I am using for my problem trainlm and trainbr training algorithms for BPNN. I am getting good results but I am confused a little bit about results of trainbr. The results produced by trainbr algorithm are better optimized than trainlm. However I need verification by some expert that either these are ok or there is any problem. The results are shown below.
>
> Input: 32, 30, 34, 35, 36
> Target: 0.9900, 0.9500, 1.0300, 1.0500, 1.0700
> Trainlm: 0.9934, 0.9523, 1.0279, 1.0536, 1.0665
> Trainbr: 0.9901, 0.9501, 1.0299, 1.0501, 1.0699
>
> These results are obtained for validation data. The results in case of trainbr are very close to the target data. However I just want to know that these results are ok or the neural network is not trained properly for trainbr. Advance thanks for any help and guidance.


Trainlm validation data is used to determine when to stop training. Therefore, it is considered part of the design data (Design = Training + Validation). Consequently, using the validation error rate to evaluate the net does not result in an unbiased estimate. Nonetheless, the validation error rate can be used to choose the "best" of multiple designs. However an unbiased evaluation must be obtained from a nondesign , test data set that is large enough to yield a robust estimate (e.g., the estimated standard deviation is much less than the estimated mean).

Trainbr does not use a validation set. Therefore, I do not really know what you are doing.

Unbiased, reasonably accurate evaluations require multiple designs using reasonable large data sets. See Wikipedia for theoretical estimates of standard deviation for
chisquare (regression) error and binomial(classification) error.

Hope tis helps

Greg



Point your RSS reader here for a feed of the latest messages in this topic.

[Privacy Policy] [Terms of Use]

© Drexel University 1994-2014. All Rights Reserved.
The Math Forum is a research and educational enterprise of the Drexel University School of Education.