Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

PRM (Predictive Rule Mining)

0 views
Skip to first unread message

Tim vor der Brück

unread,
May 15, 2008, 6:53:17 AM5/15/08
to
Hi,

I finally get along with FOIL.
Still, I heard that PRM (Predictive Rule Mining) should have superior
performance. Does someone know where to download this?
I know where a Java version is available but I would expect the
performance of a Java program to be rather poor. Thus, I am looking for
a C-version.

Best regards,

Tim

[ comp.ai is moderated ... your article may take a while to appear. ]

Ted Dunning

unread,
May 17, 2008, 9:34:30 PM5/17/08
to
On May 15, 3:53 am, Tim vor der Brueck <nos...@nospam.de> wrote:

> I know where a Java version is available but I would expect the
> performance of a Java program to be rather poor. Thus, I am looking for
> a C-version.

That assumption is unfounded. Java programs still have a moderate
penalty in terms of memory usage, but generally have little or no
penalty in terms of run-time.

Take for example Jruby, the Java implementation of Ruby. It is
generally *faster* than the C implementation.

Stefan Ram

unread,
May 17, 2008, 11:51:36 PM5/17/08
to
Ted Dunning <ted.d...@gmail.com> writes:
>That assumption is unfounded. Java programs still have a moderate
>penalty in terms of memory usage, but generally have little or no
>penalty in terms of run-time.

If you consider a factor of "1.7" (on the average of certain
benchmarks) to be "little or nothing".

"1.0 C++ GNU g++
1.7 Java 6 -server"

http://shootout.alioth.debian.org/gp4/benchmark.php?test=all&lang=all

>Take for example Jruby, the Java implementation of Ruby.
>It is generally *faster* than the C implementation.

Faster than Ruby 1.8.x, which is a pure interpreter.
Ruby 1.9 is about three times as fast.

Isaac Gouy

unread,
May 18, 2008, 11:18:50 PM5/18/08
to
On May 17, 8:51 pm, r...@zedat.fu-berlin.de (Stefan Ram) wrote:

> Ted Dunning <ted.dunn...@gmail.com> writes:
> >That assumption is unfounded. Java programs still have a moderate
> >penalty in terms of memory usage, but generally have little or no
> >penalty in terms of run-time.
>
> If you consider a factor of "1.7" (on the average of certain
> benchmarks) to be "little or nothing".
>
> "1.0 C++ GNU g++
> 1.7 Java 6 -server"
>
> http://shootout.alioth.debian.org/gp4/benchmark.php?test=all&lang=all


We know better than to accept averages at face value.

Apparently some of the Java programs are within 10% of the C++
programs, so we're left with the question - is the program I'm
interested in going to be one of those 'within 10%' Java programs, and
if not why not?

http://shootout.alioth.debian.org/gp4/benchmark.php?test=all&lang=java&lang2=gpp#ratio

> >Take for example Jruby, the Java implementation of Ruby.
> >It is generally *faster* than the C implementation.
>
> Faster than Ruby 1.8.x, which is a pure interpreter.
> Ruby 1.9 is about three times as fast.


And yet for some Ruby programs we see faster measurements on JRuby
than Ruby Core 1.9.0

http://shootout.alioth.debian.org/gp4sandbox/benchmark.php?test=all&lang=jruby&lang2=yarv#ratio

0 new messages