Re: PDP++3.0 and very large in

["Randall C. O'Reilly" <oreilly@grey.colorado.edu>]

Hans Johnson <hjohnson@mail.psychiatry.uiowa.edu> writes:
> YEAH!! Now how do I use it?  Could you please give me some hints on how
> to use this new feature?  In order to circomvent the 100,000 training
> event limit, I have broken my 5.6million training vectors into 100
> separate 56,000 training vector files (training000.pat to
> training099.pat).
> 
> I can easily concatinate these files to one large file if necessary.

go ahead and concatinate, and just edit the fromfileenv object and
point it at your file.  simplest is to use read_mode = ONE_EPOCH, set
number of events you want to consider to be one epoch (THIS IS HOW
MANY IT READS INTO RAM AT A TIME, SO DON"T MAKE IT TOO BIG!), and just
start training..

> PS:  I am writting this up as a journal paper, and would like any to
> include some appropriate references to both the design and utility of
> pdp++.  I did not see papers on your web site that specifically address
> pdp++.

unfortunately, its never been written up in a citable format.  I
suggest just citing the URL for the website.

				- Randy