Funny Junior Engine in CBLight
Wolfgang Krietsch
Hi experts!
I just discovered an extremely strange evaluation of the Junior engine when
used in CBLight:
[Event "*"]
[Site "?"]
[Date "1998.02.25"]
[Round "?"]
[White "*"]
[Black "Krietsch, Wolfgang"]
[Result "0-1"]
[SetUp "1"]
[FEN "r7/pp2k1n1/2b1p3/6q1/3P1P1r/P1PQ2P1/1P6/R4RK1 b - f3 0 35"]
[PlyCount "13"]
35... Rh1+ ({Fritz:} 35... Qd5 36. Rf3 Rh6 37. Raf1 Bb5 38. c4 Bxc4 39. Qe3
Rah8 40. Kf2 Rh2+ 41. Ke1 Qf5 42. R3f2 Qb1+ 43. Kd2 Bxf1 44. f5 Nxf5 45.
Qg5+
Kd7) 36. Kf2 Rh2+ 37. Ke1 Qh5 38. Rf2 Rxf2 39. Kxf2 Qh2+ 40. Ke3 Nf5+ 41.
Qxf5
exf5 0-1
In the variation you see what the Fritz engine thinks about it - quite ok
in my opinion, and Hiarcs comes up with a similar line.
But Junior after a couple of seconds comes up with:
35... Rh1+ 36. Kf2 Rf8 37. Ke2 Rh2 38. Ke8
... and gives an evaluation of - Mate in 8 !!!
Obviously, thsi is complete nonsense: the line Fritz/Hiarcs suggested is
much stronger, 36... Rf8 is a silly move and this line is by no way Mate in
8.
This does not happen if I analyse the same position with the Junior engine
in Fritz 5 - then Junior comes up with a similar line to one given by
Fritz/Hiarcs.
Any explanation for this?
Ciao,
Wolfgang
-------------------------------------------
REMOVE THE NUMBER FROM MY E-MAIL ADDRESS!!!
BITTE DIE NUMMER AUS DER E-MAIL ADRESSE ENTFERNEN!!!
--------------------------------------
Finger me for my PGP-Key!
39 0A ED A8 83 A7 29 0C FA 48 22 54 83 0F CF 96
Har...@t-online.de
quoting a mail from wkrie...@pobox.com
> Hi experts!
>
> I just discovered an extremely strange evaluation of the Junior engine when
> used in CBLight:
>
> [Event "*"]
> [Site "?"]
> [Date "1998.02.25"]
> [Round "?"]
> [White "*"]
> [Black "Krietsch, Wolfgang"]
> [Result "0-1"]
> [SetUp "1"]
> [FEN "r7/pp2k1n1/2b1p3/6q1/3P1P1r/P1PQ2P1/1P6/R4RK1 b - f3 0 35"]
> [PlyCount "13"]
>
> But Junior after a couple of seconds comes up with:
>
> 35... Rh1+ 36. Kf2 Rf8 37. Ke2 Rh2 38. Ke8
>
> ... and gives an evaluation of - Mate in 8 !!!!
>
> Obviously, thsi is complete nonsense: the line Fritz/Hiarcs suggested is
> much stronger, 36... Rf8 is a silly move and this line is by no way Mate in
> 8.
> This does not happen if I analyse the same position with the Junior engine
> in Fritz 5 - then Junior comes up with a similar line to one given by
> Fritz/Hiarcs.
> Any explanation for this?
According to two other positions where Junior shows strange moves and the
answer of Amir, it is not important what is shown in the window but played
on the board.
Harald Faber
Wolfgang Krietsch
On 28 Feb 1998 09:05:14 GMT, Har...@t-online.de wrote:
>
>According to two other positions where Junior shows strange moves and the
>answer of Amir, it is not important what is shown in the window but played
>on the board.
>
Sounds funny... when I analyse something in CBLight, I can only use the
suggestions made in the analysis window - and if it's not obviously
nonsense like the example I posted, I will possibly end up with a
completely misleading analysis. After all, that's what I use it for:
analysing positions I don't understand myself!!!
Maybe Amir could clarify this...
Har...@t-online.de
quoting a mail from wkrie...@pobox.com
> >According to two other positions where Junior shows strange moves and the
> >answer of Amir, it is not important what is shown in the window but played
> >on the board.
> Sounds funny... when I analyse something in CBLight, I can only use the
> suggestions made in the analysis window - and if it's not obviously
> nonsense like the example I posted, I will possibly end up with a
> completely misleading analysis. After all, that's what I use it for:
> analysing positions I don't understand myself!!!
> Maybe Amir could clarify this...
He already did...said, important is the move he makes, not the complete PV
or the investigated moves...
Harald Faber
Komputer Korner
While we are on the subject of Junior, It would be very illuminating if Amir
once and for all would tell us what the ply number means. It can't mean full
width and it can't mean full width /2 nor full width /3 nor full width/4. So
just what does it signify? Users like to know how many full width plies have
been calculated. And every other program with the exception of CM5500 gives
us a true picture of the full width plies. Amir wouldn't be giving out any
trade secrets by telling us.
--
- -
Komputer Korner
The inkompetent komputer
If you see a 1 in my email address, take it out before replying.
Please do not email both me and the r.g.c.c. at the same time. I read all
the postings on r.g.c.c.
Also every statement of mine should be taken with a grain of salt. Read at
your own risk and
assume that it is only this humble komputer's opinion.
Har...@t-online.de wrote in message <6ddni8$tql$1...@news01.btx.dtag.de>...
bruce moreland
>And every other program with the exception of CM5500 gives
>us a true picture of the full width plies.
Actually they all lie, but they may lie somewhat consistently with
each other.
If you do a min-max search to depth D, you can say you searched
through depth D, because you did.
If you do an alpha-beta search to depth D, same thing, since
alpha-beta will return the same result as min-max, with a few
ignorable exceptions due to transposition table effects.
If you use any sort of pruning (including null-move forward pruning),
it is less honest to say you get through depth D, since you'll miss
things that someone who does pure alpha-beta won't miss.
Likewise with extensions. An extension allows you to see beyond the
depth you normally get.
Also, programs can vary in how they choose to deal with the situation
once then run out of draft -- some do a more complex quiescent search
than others.
bruce
Wolfgang Krietsch
On 2 Mar 1998 07:36:40 GMT, Har...@t-online.de wrote:
>
>quoting a mail from wkrie...@pobox.com
>
>
>> Sounds funny... when I analyse something in CBLight, I can only use the
>> suggestions made in the analysis window - and if it's not obviously
>> nonsense like the example I posted, I will possibly end up with a
>> completely misleading analysis. After all, that's what I use it for:
>> analysing positions I don't understand myself!!!
>> Maybe Amir could clarify this...
>
>He already did...said, important is the move he makes, not the complete PV
>or the investigated moves...
>
That means that Junior's evaluation value is worthless? I'm not sure, but I
think KK stated that Junior has the most accurate evaluation values of all
chess programs... giving the "Mate in 8" evaluation of the prevous
mentioned position.... hmmmmmm
Amir Ban
>
>
> While we are on the subject of Junior, It would be very illuminating if Amir
> once and for all would tell us what the ply number means. It can't mean full
> width and it can't mean full width /2 nor full width /3 nor full width/4. So
> just what does it signify? Users like to know how many full width plies have
> been calculated. And every other program with the exception of CM5500 gives
> us a true picture of the full width plies. Amir wouldn't be giving out any
> trade secrets by telling us.
>
> --
> - -
> Komputer Korner
Once and for all ? I doubt there's anything I can say to settle this.
There's no excuse for doing things differently :) I've never been very
secretive about this (and it's not a big secret). I answered this in the
newsgroups, in Computer Chess Reports, and in conversations.
The brute-force ply depth is indeed half the publicized depth. All the
rest are extensions (in conventional terminology, I don't think of them
this way). If you set Junior to depth 12, e.g., then you should be able
to find a 7-ply combination where it fails. If I am doing a good job,
then you should have a hard time finding one.
The question of what this is equivalent to in terms of other programs,
e.g. a null-mover with "standard" extensions is interesting, but I don't
know the answer. In tournament conditions middlegame Junior typically
gets 14-16 depths, and it looks competitive tactically.
Amir
Chris Whittington
Komputer Korner <kor...@netcom.ca> wrote in article
<6derhc$an8$1...@tor-nn1.netcom.ca>...
> While we are on the subject of Junior, It would be very illuminating if
Amir
> once and for all would tell us what the ply number means. It can't mean
full
> width and it can't mean full width /2 nor full width /3 nor full width/4.
So
> just what does it signify? Users like to know how many full width plies
have
> been calculated. And every other program with the exception of CM5500
gives
> us a true picture of the full width plies.
Mine doesn't. I don't actually see how it could; since I don't know the
answer myself.
This is more complex than you imagine, and can't be reduced to trivial
numbers such as 'full width ply depth'
Chris Whittington
> Amir wouldn't be giving out any
> trade secrets by telling us.
>
> --
> - -
> Komputer Korner
>
> The inkompetent komputer
>
> If you see a 1 in my email address, take it out before replying.
> Please do not email both me and the r.g.c.c. at the same time. I read all
> the postings on r.g.c.c.
> Also every statement of mine should be taken with a grain of salt. Read
at
> your own risk and
> assume that it is only this humble komputer's opinion.
> Har...@t-online.de wrote in message <6ddni8$tql$1...@news01.btx.dtag.de>...
> >
> >quoting a mail from wkrie...@pobox.com
> >
> >
> >> >According to two other positions where Junior shows strange moves and
> the
> >> >answer of Amir, it is not important what is shown in the window but
> played
> >> >on the board.
> >
> >> Sounds funny... when I analyse something in CBLight, I can only use
the
> >> suggestions made in the analysis window - and if it's not obviously
> >> nonsense like the example I posted, I will possibly end up with a
> >> completely misleading analysis. After all, that's what I use it for:
> >> analysing positions I don't understand myself!!!
> >> Maybe Amir could clarify this...
> >
> >He already did...said, important is the move he makes, not the complete
PV
> >or the investigated moves...
> >
> >
> >Harald Faber
> >
>
>
>
Har...@t-online.de
quoting a mail from wkrie...@pobox.com
> >> Sounds funny... when I analyse something in CBLight, I can only use the
> >> suggestions made in the analysis window - and if it's not obviously
> >> nonsense like the example I posted, I will possibly end up with a
> >> completely misleading analysis. After all, that's what I use it for:
> >> analysing positions I don't understand myself!!!
> >> Maybe Amir could clarify this...
> >
> >He already did...said, important is the move he makes, not the complete PV
> >or the investigated moves...
> That means that Junior's evaluation value is worthless?
No! The PV is obviously buggy, also some moves Junior thinks about. but
the eval seems to be OK.
> I'm not sure, but I
> think KK stated that Junior has the most accurate evaluation values of all
> chess programs... giving the "Mate in 8" evaluation of the prevous
> mentioned position.... hmmmmmm
Don't know, try out position #2 in ComputerSchach&Spiele, until depth 40
no mate announced, Hiarcs6 and Fritz5 are much faster (in about 2mins K6-
200, 44MB hash) although they count the mate in xx wrong. :-)
Harald Faber
Komputer Korner
Thanks Amir, but if what you say is true, then Junior is a very selective
search program because I have seen it search for 30 minutes and only get to
15 plies, which is only 7 and one half plies in your terminology. Which is
only 3 and 3/4 full moves, which is not very deep in full width terms.
--
- -
Komputer Korner
The inkompetent komputer
If you see a 1 in my email address, take it out before replying.
Please do not email both me and the r.g.c.c. at the same time. I read all
the postings on r.g.c.c.
Also every statement of mine should be taken with a grain of salt. Read at
your own risk and
assume that it is only this humble komputer's opinion.
Amir Ban wrote in message <34FBE9...@m-sys.com>...
Robert Hyatt
Komputer Korner <kor...@netcom.ca> wrote:
: Thanks Amir, but if what you say is true, then Junior is a very selective
: search program because I have seen it search for 30 minutes and only get to
: 15 plies, which is only 7 and one half plies in your terminology. Which is
: only 3 and 3/4 full moves, which is not very deep in full width terms.
Of course, check out Genius. where 8 takes a good bit of time. And if
you compare that to Fritz, don't. Fritz (like Crafty) uses null-move
and a 14 ply null-move search can overlook things that a 10 ply full-width
search won't, due to the way it prunes.
In short, comparing search depth is about like comparing evaluations
which is about like comparing two people's belly-buttons. Interesting
but little useful information comes out of it...
: --
: - -
: Komputer Korner
: The inkompetent komputer
: If you see a 1 in my email address, take it out before replying.
: Please do not email both me and the r.g.c.c. at the same time. I read all
: the postings on r.g.c.c.
: Also every statement of mine should be taken with a grain of salt. Read at
: your own risk and
: assume that it is only this humble komputer's opinion.
: Amir Ban wrote in message <34FBE9...@m-sys.com>...
:>Once and for all ? I doubt there's anything I can say to settle this.
:>There's no excuse for doing things differently :) I've never been very
:>secretive about this (and it's not a big secret). I answered this in the
:>newsgroups, in Computer Chess Reports, and in conversations.
:>
:>The brute-force ply depth is indeed half the publicized depth. All the
:>rest are extensions (in conventional terminology, I don't think of them
:>this way). If you set Junior to depth 12, e.g., then you should be able
:>to find a 7-ply combination where it fails. If I am doing a good job,
:>then you should have a hard time finding one.
:>
:>The question of what this is equivalent to in terms of other programs,
:>e.g. a null-mover with "standard" extensions is interesting, but I don't
:>know the answer. In tournament conditions middlegame Junior typically
:>gets 14-16 depths, and it looks competitive tactically.
:>
:>Amir
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170
Komputer Korner
I didn't say that the moves were the best. I said that the Junior evaluation
score seemed to represent very well the positional value of compensation in
a position. Junior seems to have a very good score algorithm. This has
nothing to do with how good it plays because it is one of the most selective
searchers on the market. Don't trust any of Junior's actual moves unless you
have left the position on all night. The 2 concepts above are quite
different things. many programmers will tell you not to look at their
program's evaluation score at all or take it with a grain of salt. I like
the way Junior evaluates a position. This however may or may not correspond
to playing good chess.
--
- -
Komputer Korner
The inkompetent komputer
If you see a 1 in my email address, take it out before replying.
Please do not email both me and the r.g.c.c. at the same time. I read all
the postings on r.g.c.c.
Also every statement of mine should be taken with a grain of salt. Read at
your own risk and
assume that it is only this humble komputer's opinion.
Wolfgang Krietsch wrote in message <34fb95b1...@news.cityweb.de>...
>On 2 Mar 1998 07:36:40 GMT, Har...@t-online.de wrote:
>
>>
>>He already did...said, important is the move he makes, not the complete PV
>>or the investigated moves...
>>
>That means that Junior's evaluation value is worthless? I'm not sure, but I
Wolfgang Krietsch
On Thu, 5 Mar 1998 00:12:46 -0500, "Komputer Korner" <kor...@netcom.ca>
wrote:
>I didn't say that the moves were the best. I said that the Junior evaluation
>score seemed to represent very well the positional value of compensation in
>a position. Junior seems to have a very good score algorithm. This has
That's what I was mainly talking about also. Just to remember:
> [Event "*"]
> [Site "?"]
> [Date "1998.02.25"]
> [Round "?"]
> [White "*"]
> [Black "Krietsch, Wolfgang"]
> [Result "0-1"]
> [SetUp "1"]
> [FEN "r7/pp2k1n1/2b1p3/6q1/3P1P1r/P1PQ2P1/1P6/R4RK1 b - f3 0 35"]
> [PlyCount "13"]
>
> But Junior after a couple of seconds comes up with:
>
> 35... Rh1+ 36. Kf2 Rf8 37. Ke2 Rh2 38. Ke8
>
> ... and gives an evaluation of - Mate in 8 !!!!
It's this evaluation, which is obviously nonsense, what hit me...
Amir Ban
Wolfgang Krietsch wrote:
>
> > [FEN "r7/pp2k1n1/2b1p3/6q1/3P1P1r/P1PQ2P1/1P6/R4RK1 b - f3 0 35"]
> > But Junior after a couple of seconds comes up with:
> >
> > 35... Rh1+ 36. Kf2 Rf8 37. Ke2 Rh2 38. Ke8
> >
> > ... and gives an evaluation of - Mate in 8 !!!!
>
> It's this evaluation, which is obviously nonsense, what hit me...
>
> Ciao,
> Wolfgang
There's a Junior bug in interpreting the en-passant field in the FEN.
The Junior line would be correct if the pawn on f4 is removed. Remove
the en-passant field thus:
[FEN "r7/pp2k1n1/2b1p3/6q1/3P1P1r/P1PQ2P1/1P6/R4RK1 b - - 0 35"]
and you will see correct behavior.
Amir
Wolfgang Krietsch
On Thu, 05 Mar 1998 10:15:18 +0200, Amir Ban <ami...@m-sys.com> wrote:
>
>There's a Junior bug in interpreting the en-passant field in the FEN.
>The Junior line would be correct if the pawn on f4 is removed. Remove
>the en-passant field thus:
>
>[FEN "r7/pp2k1n1/2b1p3/6q1/3P1P1r/P1PQ2P1/1P6/R4RK1 b - - 0 35"]
>
>and you will see correct behavior.
That's it! Thanks!
Funny, though - since there is nothing that could be taken en-passant
anyway...
mclane
Robert Hyatt <hy...@cis.uab.edu> wrote:
>In short, comparing search depth is about like comparing evaluations
>which is about like comparing two people's belly-buttons. Interesting
>but little useful information comes out of it...
The guy speaks my language ! Good examples Bob !