Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Toga mauls Shredder

Author: Stephen Ham

Date: 09:40:50 09/01/05

Go up one level in this thread


On September 01, 2005 at 11:20:51, Greg Simpson wrote:

>The source is available for both, they are not entirely the same.
>
>I tried to see whether Toga 1.0beta2 was better head to head against Fruit 2.1
>in two minute/40 games, but after hundreds of games I couldn't reach a reliable
>conclusion.  Toga II 1.0 quickly showed itself to be better.
>
>Of course, the unreleased version of Fruit could well be better still.

Dear Gentlemen,

As a reader who's ignorant regarding chess programming, I was shocked and amazed
that these amateur programs, Fruit and Toga can be the equals, if not superiors
to Shredder 9, a commercial program that dominates all other commercials.

But I've downloaded both Fruit 2.1, and subsequently Toga II, and was astounded
by the results.

I began by creating a double round-robin tournament with extremely long time
controls (e.g. 6-hours/40-moves, 2-hours for the next time control, and then
1-hour to finish) on a very fast AMD computer with 708 MB's of RAM. It consisted
of Shredder 9, Junior 9, Hiarcs 9 and Fruit 2.1. My Fruit didn't have an opening
book, so I gave it one that I created to test my correspondence chess opening
ideas/novelties (lots of Dragons and Gruenfelds for Black, and my 1 d4 opening
lines for White).

The result was a clear win for Fruit, followed by Shredder, then Junior, and
finally Hiarcs. I noticed that Fruit, while very strong, doesn't play sharply
and that my opening book may not have been an ideal match for it. So I gave it
the Nimzo 7.32 opening book for the next tourney.

By this time, I'd downloaded Toga II and gave it my correspondence chess book.
My next tourney was another double round-robin with classic FIDE time controls
and the order of the finish was: Shredder, Toga, Junior, Fruit, and Hiarcs
(scoring only 1.5!).

I still see this as a great result for Fruit and Toga. But, I think that Fruit's
use of the older Nimzo book may not have been ideal, since it didn't like many
of the openings it got. So I "upgraded" to a modified Fritz 7 book (modified
with data from recent opening theory and corrections of some book errors). I
felt this book was solid enough to match Fruit's sold style. So in my last
double round-robin tourney at standard time controls, the result was: Shredder,
Fruit and Toga just a 1/2-point behind, Junior, and Hiarcs.

Some observations from the above minimal data:
1) Fruit showed great strength in the endgames.

2) It also has a realistic evaluation function that's much superior to
Shredder's. While it doesn't generally engage in aggressive play like Junior or
Shredder (Toga seemed sharp too), it's not tactically weak at all.

3) Toga seemed more variable in performance, while Fruit was universally
steady/solid. It's my perception, based upon too little data, that Toga is a
fine tactician but seems a little weaker in the endgame than Fruit (again based
upon just these games).

4) Hiarcs is a great program, but it was hurt by some awful stuff in its opening
book. I've been gradually improving its book lines as I come across faulty
lines. But clearly there's much more work to do. Hiarcs performs much better
when I give it my modified Fritz 7 book. But I'll never get the Hiarcs book
fixed unless I play more with it. Hiarcs was thus hurt by beginning games
against superior engines, at long time-controls, from generally poor starting
positions after its book ended. For the record, the bad Hiarcs book actually
seems to help Hiarcs in speed games though, since Hiarcs scores very well (often
coming in first) in my few short time-control tourneys. I've yet to test Fruit
and Toga at fast time controls, since I'm really not interested in these
results.

I've subsequently tried using Fruit to test my opening novelties, but it doesn't
perform the "Deep Position Analysis" properly. For example, if it's White's turn
to move after 15-moves, it begins its search by skipping White's move and then
looks only at the position as if Black were to move. It also fails to display
the options other than the PV.

Anyway, I'll test further. But I'm impressed. So while I see no dominance over
Shredder's results in my tests, it's indeed possible that Toga and Fruit are
superior to Shredder at long time controls.

Again, these are just the observations of a computer dummy who knows a thing or
two about chess in general, based upon too few games at long time controls.

All the best,

Steve



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.