
Is Emping an implementation of C4.5 or something new? No, it's new. The reduction algorithm is not based on a measure and is not a search either. It's a construction of predicate combinations, starting with singletons, that are not contradictory to known facts. The
On Mon, 2007-05-21 at 14:06 -0700, Scott Cruzen wrote: third step, following hypothesis and then falsification, is verification with observed rules. The complexity appears to be quadratic to the original rule length and also to the number of original rules, but I have yet to prove this. I believe I have proved that the algorithm produces the complete and shortest disjunctive normal form of a logical rule table. The white paper 'Deriving heuristic rules from facts' http://j-van-thiel.speedlinq.nl/dwnl/EmpingWP.pdf (21 pages) discusses the theory from examples, and an earlier wp (also available on the web site) uses some first order predicate logic and set theory. Version 0.2 also shows logical dependencies in the reduced rules and this is just a search in the original rules and in the poset of reductions. So this is not new in a theoretical sense, though I believe no other implementation has ever considered such dependencies. Regards, Hans van Thiel
* Hans van Thiel
[070521 05:45]: Hello All,
Version 0.2 of Emping, a utility to derive heuristic rules from a table of nominal data, is available. In addition to a reduced normal form in .csv format, which can be read by Open Office Calc, it now also displays observed implications and equivalences in the reduced rules (if present) and ambiguities in original rules (if present). A bug in the response to a user typing error has been fixed and some code has been simplified, compared to 0.1. An online user guide is at: http://j-van-thiel.speedlinq.nl/emp/empug.html and the Cabal package can be downloaded from there, or from the HackageDB.
Thanks,
Hans van Thiel
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
participants (1)
-
Hans van Thiel