Re: cascade correlation in batch mode

(Që…ð@ÄØð@ÄÔ)
Mon, 28 Jul 1997 14:06:22 +0200 

>>>>> "Theo" == Theo Sabisch <elerts@herts.ac.uk> writes:

|> hi,

|> i am trying desperately to use the cascade correlation algorithm in
|> the batch interpreter of snns v4.2.

|> to run the algorithm correctly you have to specify quite a number
|> of parameters which normally are entered into the snns cascade
|> panel, right?  which function takes all these arguments in the
|> batch interpreter?  is it the learning function?  do i have to
|> enter them in exactly the same order they appear in the panel?  i'm
|> really a bit confused about that.  any help would be greatly
|> appreciated.

To use the CC algorithms you have to use the function calls 


setCascadeParams(...)Set the additional parameters required for training 
                   a cascade correleation network. The parameters are:

                   max outp. unit error (float)
                   subordinate learning function ("Quickprop","Backprop",or "Rprop")
                   print covar. and error (TRUE or FALSE)
                   prune new hidden unit (TRUE or FALSE)
                   minimization function ("SBC" , "AIC", or "CMSEP")
                   min. covar. change (float)
                   cand. patience (int)
                   max no of covar. updates (int)
                   max no of candidate units (int)
                   activation function ("Act_Logistic", "Act_LogSym",
                                        "Act_TanH", or "Act_Random")
                   error change (float)
                   output patience (int)
                   max no of epochs (int)

                   For a detailed description of these parameters see
                   the pruning section of the manual. As usuall with
                   batchman, latter parameters may be skipped if the
                   default values are to be taken. Defaults would be:
                   0.2, "Quickprop", TRUE, FALSE, "SBC", 0.04, 25, 200, 8,
                   "Act_LogSym", 0.01, 50, 200
                   
delCandUnits()     delete all the candidate units created by CC.


Unfortunately they are NOT documented in the current version of the
manual!

Guenter
 -------------------------------------------------------------------------
|Guenter W. Mamier                      mamier@informatik.uni-stuttgart.de|
|University of Stuttgart, IPVR		       	                          |
|Breitwiesenstrasse 20-22, 						  |
|70565 Stuttgart, Germany                            Tel: +49 711 7816-447|
|                                                    Fax: +49 711 7816-250|
| http://www.informatik.uni-stuttgart.de/ipvr/bv/personen/mamier_engl.html|
 -------------------------------------------------------------------------