Surprising results on image recognition?!

Steffen Nissen

2007-01-20 20:35:45 UTC

Sorry, think that I misunderstood you on this one, thought that you
compared it to new data, not test data.

If your problem is a classification problem, then there should be a
direct correlation with the bit fail. If you set the bit fail_limit to
the same as your succes criteria, then the bit fail should be exactly
the same as your succes rate.

Hope that this helps,
Steffen

Thank you a lot Steffen.
Would you also suggest the reason why there is no direct relationship
between MSE (of test data), Bit Fail and success ratio (of test data)?!
How can this happen? Or am I doing something wrong?
Thank you again,
Best Wishes,
Berk Anbaroglu
Hi Berk,
On the question about succes ratio, then this can differ alot, since
two networks can get the same MSE, but if the underlying ANN is
different, then the results with new data might not be the same.
And yes, the learning rate has no effect on RPROP, RPROP has it's own
set of parameters.
Steffen

Dear Steffen Nissen,
* I trained exactly the same ANN several times; and each time I got a
different success ratio; ranging from 27,6% to %42.
* I couldn't see a relationship between MSE, bit fail and success ratio!
(In ANN's best performance, it had, MSE (on test data) of 0.005232 and Bit
fail of 180; however there are less MSE and Bit Fail results on other
trainings)
* Finally is it true that; learning rate has no effect on RPROP?
(Because in the best performance, stated above, I set the learning rate to
0.2)
And I couldn't see the reason of these 3 points!
If you have any idea, please let me know.
Thank you for your interest!
Best Wishes,
Berk Anbaroglu
________________________________
Never miss an email again!
Yahoo! Toolbar alerts you the instant new Mail arrives. Check it out.

--
Steffen Nissen - http://MySpace.com/SteffenNissen
Project Administrator - Fast Artificial Neural Network Library (fann)
http://fann.sf.net
________________________________
Don't pick lemons.
See all the new 2007 cars at Yahoo! Autos.

--
Steffen Nissen - http://MySpace.com/SteffenNissen
Project Administrator - Fast Artificial Neural Network Library (fann)
http://fann.sf.net

-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV