trainbr - question about Effective # Param (gamk)

Hello everybody! At some moment of training, Effective # Param (gamk) begins highly jittering, and sometimes take negative values. Is that normal?

[Greg Heath]

I have no idea.

Do you get glitches if you use trainlm or trainscg?

How large is your data set?

Care to share? If data set is huge not necessary to send all.

*.txt or *.m format preferred

Greg

[Arekusandoru]

Sorry for the late response - I have missed email notification.

• When I use trainlm there is no glitches, so I use it now.
• Data set is above thousands of samples. From several K to 20-30K, depends on the current experiment.
• I will check this glitch on the weekend and if it is still present I prepare for you *.m files (if that is still relevant for you)

Answer 1

[Greg Heath]

Check the source code of trainbr.

 type trainbr

You will probably find if statements that discontinuously change one or more training parameters.

Hope this helps.

Thank you for formally accepting my answer

Greg