Horvath Play Styles

This forum is for general discussions and questions, including Collectors Corner and anything to do with Computer chess.

Moderators: Harvey Williamson, Steve B, Watchman

Forum rules
This textbox is used to restore diagrams posted with the fen tag before the upgrade.
User avatar
paulwise3
Senior Member
Posts: 1505
Joined: Tue Jan 06, 2015 10:56 am
Location: Eindhoven, Netherlands

Post by paulwise3 »

A few months ago I played a small testmatch between the Concerto (normal style) and the Mephisto Europa. Because of the Europa's active playing style the Concerto had a lot of trouble, and in four games it became a 2-2 draw: one loss, one win and two draws. And one of those draws should have been a win for the Europa (stupid move repetition).
(Edit: on the other hand, the second draw should have been a win for the Concerto, it wasted a totally won position).
Because of my idea that with all feature values set to 20 it played more active I started a new testmatch. And now the Concerto acted like a professional matchplayer: 3 wins with white, and 3 quick draws with black by move repetition. Apparently the Europa has no check for this, because in all three draws it had a slight advantage.
Anyway, this score matches better with the 100 elo-points difference on the Wiki-elo-liste.

Considering new tests with all feature values set to 30(/40/50/60/70/80/90) regards,
Paul
User avatar
paulwise3
Senior Member
Posts: 1505
Joined: Tue Jan 06, 2015 10:56 am
Location: Eindhoven, Netherlands

Post by paulwise3 »

The Dutch Computer Chess federation (CSVN.nl) organizes its 30th chesscomputer user tournament on June 13 in Leiden. The wellknown Dutch master Hans Böhm will play a simultan against 12 chesscomputers then. I subscribed with the Sphinx Concerto. It will very probably loose, but I am looking for optimal feature values for it to play with black.
So far it looks like the default settings (playing style 5) are recommended. With all feature values set to 20 I had some nice results with white, but I am not sure it is good for playing with black.
Anybody (especially Nick!?) have suggestions?

Concerto smelling blood regards,
Paul.
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

paulwise3 wrote:The Dutch Computer Chess federation (CSVN.nl) organizes its 30th chesscomputer user tournament on June 13 in Leiden. The wellknown Dutch master Hans Böhm will play a simultan against 12 chesscomputers then. I subscribed with the Sphinx Concerto. It will very probably loose, but I am looking for optimal feature values for it to play with black.
So far it looks like the default settings (playing style 5) are recommended. With all feature values set to 20 I had some nice results with white, but I am not sure it is good for playing with black.
Anybody (especially Nick!?) have suggestions?

Concerto smelling blood regards,
Paul.
Hi Paul,

That should be a fun experience for you. Good luck and have fun! I would stick with the standard setting. The few tests that I had done indicate that the standard setting seems to be the best with Black.

ps. I had been thinking about taking some computers to the local chess club here where I live and see if I can get some of the rated players to play them. Maybe someday I will do that.

Best regards
Nick
User avatar
paulwise3
Senior Member
Posts: 1505
Joined: Tue Jan 06, 2015 10:56 am
Location: Eindhoven, Netherlands

Post by paulwise3 »

spacious_mind wrote: Hi Paul,

That should be a fun experience for you. Good luck and have fun!
Hi Nick,

Thanx, and yes I'm looking forward to it :-)
Still, I'll try to experiment a little with these features
I will keep you informed, also about the outcoming of the simultan match.

Cannot wait regards,
Paul.
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

I decided to play some 1/2 second test games to see how a computer would score and compared it to how they scored in the 30 seconds per move test. I used Test game 5 to do the comparison. And of course I thought I would use the computers which are the topic of the day.

Image

Amazingly they all scored much better then I expected at Level 7 (1-2 sec per move) At the bottom of the chart you can see the rating improvement between Level 7 and Level 53.

Also I listed the move deviations of each computer for both Level 7 and level 53. As you can see from the chart there is almost as much move deviation between Regency over Legend and the Excalibur's over Legend. In fact at level 7 Regency had less deviation with Ivan then it did with Legend. But the table also confirms a closer relationship between Ivan and Igor.

Personally my opinion remains unchanged. I still think that someone used a low level setting and configured the computers to a point where they felt that they played their best move (in their mind). Hence the interesting performances of Regency, Igor and Ivan at level 7. However as the time per move increases this improvement gradually gets eaten away again.

My opinion still remains that I believe that someone who had access to the configuration tables played around with the configurations until they were satisfied that they had something that they liked. I still believe the root author of the programs remains as Horvath.

But that is just my opinion therefore don't get mad and don't get personal.

Best regards
Nick
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

With Fernando's suggestion to Test some Spracklen's here is the 1st test with Fidelity Super 9. Super 9 has a 6502 2 MHz program. I am guessing but I would say that the H8 10/12 MHZ programs are about 3 times faster.

Image

Definitely in this test Super 9 had fewer different moves when comparing to Igor and Ivan than comparing to Legend and Regency. But still 23 (Igor) and 25 (Ivan) deviations is still a lot considering that the total amount of moves is 46 which makes it 50% same Igor and 46% same Ivan.

But there is one move that both the Excalibur's and Super 9 have in common that is just an incredibly bad move:

[fen]rnb1kb1r/3p1pp1/2p2q1p/pp2p3/4P3/1BNP4/PPP2PPP/R2QK1NR w KQkq - 0 8[/fen]

In the above position Igor/Ivan and Super 9 will play 8. Qf3? Which of course is just really really bad since you lose the Bishop and the game before you even started to play the game. The two Horvath's have not shown this move in the above tests.

But on the flip side if you look at the chart at Regency black moves 16, 17 & 18 and Super 9 black moves 17, 18 and 19. You see the tendency for pointless if not exactly the same move played.

Other than 8. Qf3? I have not seen any other distinguishing moves to point towards a Spracklen engine with Super 9. The 5 seconds per move should be fairly close to the 1/2 seconds per move in the first test. I might later try to play a higher level than 30 seconds as well with Super 9 to see if there is a closer match.

Best regards
Nick
User avatar
Steve B
Site Admin
Posts: 10140
Joined: Sun Jul 29, 2007 10:02 am
Location: New York City USofA
Contact:

Post by Steve B »

spacious_mind wrote:

But there is one move that both the Excalibur's and Super 9 have in common that is just an incredibly bad move:

[fen]rnb1kb1r/3p1pp1/2p2q1p/pp2p3/4P3/1BNP4/PPP2PPP/R2QK1NR w KQkq - 0 8[/fen]

In the above position Igor/Ivan and Super 9 will play 8. Qf3? Which of course is just really really bad since you lose the Bishop and the game before you even started to play the game. The two Horvath's have not shown this move in the above tests.


Other than 8. Qf3? I have not seen any other distinguishing moves to point towards a Spracklen engine with Super 9. The 5 seconds per move should be fairly close to the 1/2 seconds per move in the first test. I might later try to play a higher level than 30 seconds as well with Super 9 to see if there is a closer match.

Best regards
Just for your info Nick
a question arose in another thread about the Mephisto Explorer
as we know the computer is a Morsch program

anyway out of curiosity and since i has the computer already out..i set up this position and guess what..
it plays 8.Qf3 at the 5 second level

I'm Just Sayin Regards
Steve
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Steve B wrote:
spacious_mind wrote:

But there is one move that both the Excalibur's and Super 9 have in common that is just an incredibly bad move:

[fen]rnb1kb1r/3p1pp1/2p2q1p/pp2p3/4P3/1BNP4/PPP2PPP/R2QK1NR w KQkq - 0 8[/fen]

In the above position Igor/Ivan and Super 9 will play 8. Qf3? Which of course is just really really bad since you lose the Bishop and the game before you even started to play the game. The two Horvath's have not shown this move in the above tests.


Other than 8. Qf3? I have not seen any other distinguishing moves to point towards a Spracklen engine with Super 9. The 5 seconds per move should be fairly close to the 1/2 seconds per move in the first test. I might later try to play a higher level than 30 seconds as well with Super 9 to see if there is a closer match.

Best regards
Just for your info Nick
a question arose in another thread about the Mephisto Explorer
as we know the computer is a Morsch program

anyway out of curiosity and since i has the computer already out..i set up this position and guess what..
it plays 8.Qf3 at the 5 second level

I'm Just Sayin Regards
Steve
Thanks Steve, good to know. So Morsch plays that horrible move as well! :) Can you check when you got a moment if it switches to something else at 30 seconds level. Igor does switch it. But Ivan keeps it as do the couple of Spracklens I have tried so far.

Best regards
Nick
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

You are right Steve. Not the best example to pin my case on. Will have to play them through all the tests and see if better matches come up:

Image

From the 90 or so programs tested so far, the above all play 8. Qf3 after 30 seconds.

That makes it for sure Spracklen does with both Super 9, & Super 9 Deluxe (just finishing the Test game with SU9 Deluxe. Morsch, Kaplan, Kittinger & compies in question. Forget the Danielsen mention for now. I have assumed it would be the same as Lexibook's Glass Chess but maybe it is not. If someone has Lexibook Glass Chess then please play the test if you have time.

Thanks
Nick
Nick
User avatar
Steve B
Site Admin
Posts: 10140
Joined: Sun Jul 29, 2007 10:02 am
Location: New York City USofA
Contact:

Post by Steve B »

Sure Nick
at 30 sec /Avg. Explorer will provide an escape square for it B with 8.a3

Hemmed In Regards
Steve
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

Steve B wrote:Sure Nick
at 30 sec /Avg. Explorer will provide an escape square for it B with 8.a3

Hemmed In Regards
Steve
Thanks, it would have been a shocker for Morsch to play 8. Qf3 at 30 seconds.

I just updated the table in the previous post to show both SU9 and SU9 Deluxe.

Best regards
Nick
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

I tried to do some more comparisons to look for any similarities. I also added Peri Beta to the comparison. Not really a very good choice but I just wanted to see closer what it would look like playing at same game level setting of 30 seconds. I tried Sensory Voice but it is a lot trickier because it does not have a take back function. Therefore you have to do set up after each move. At Level CL3 (35 seconds) by doing it this way, it had a tendency to play its move after about 10 seconds so I gave up on it. I might have to do it with a timer someday and use analysis level to see if it works better and force the move instead.

MORE TEST GAME 5 COMPARISONS

Image

This test if you look at the bottom score, does show SU9's score leaning closer to the Excalibur computers than the two Horvath's, with Peri Beta the furthest away in similarity to Excalibur.

I am going to add some faster Spracklen later but wanted to start with some good older Spracklen programs first.

TEST GAME 5 UNIQUE MOVES

In an earlier post I showed that the moves 8. Qf3 followed by 9. Qf3 are quite unique. But there some other chess programs that also play this. But the sequence with Black moves 26. ... Nh4+ followed by 27. ... Nh4+ is unique from all the tested computers. So far only Excalibur & Spracklen have played these.

Image

These are the only computers from all tested so far that play the 26. ... Nh4+ followed by 27. ... Nh4+.

TEST GAME 3 COMPARISONS

Image

Test game 3 really does not have any unique moves that stand out, but without checking you just don't know. But again the SU9's show to be marginally closest to Excalibur than Legend. Furthest away again is Regency followed by Peri Beta.


TEST GAME 2 COMPARISONS

Test game 2 is interesting because for this test I also have Excalibur Avenger and Grandmaster Platinum. Steve B. had kindly supplied these test scores. And, also some scores of other tests that I had done.

Image

You can immediately see that Regency and Avenger are probably the same Horvath program. Which makes me think that there really are 3 version all named around Legend. The slight move difference within each version could be just a slight difference in hardware:

Legend = Concerto = Accolade
Legend 2 = Challenge
Legend 3 = Avenger = Regency

It could be that simple. Same could apply also to the small variations between Mirage, Ivan, Igor & Grandmaster.

You can also see in this test game that the Excalibur computers sit somewhere around the middle between Horvath's and Spracklen with Peri Beta being very far away in this test comparison.

TEST GAME 2 UNIQUE MOVES

Test Game 2 also has a unique sequence of two consecutive bad moves with 24. Qb3 & 25. Qb2.

Image

Kaplan with Maestro B is the one exception in the above chart. Excalibur plays this sequence, Spracklen plays this sequence. And, there are some more Excalibur additions that play this sequence.

Image

Kittinger plays it with Novag Aquamarine Milenio. Alexandra plays it, and more RadioShack/Excalibur and Ivan/Igor/Mirage also play it at the fixed time level setting 22/23.

Image

Lastly Kaare Danielsen plays it!

Best regards
Nick
User avatar
paulwise3
Senior Member
Posts: 1505
Joined: Tue Jan 06, 2015 10:56 am
Location: Eindhoven, Netherlands

Post by paulwise3 »

Steve B wrote:
spacious_mind wrote:

But there is one move that both the Excalibur's and Super 9 have in common that is just an incredibly bad move:

[fen]rnb1kb1r/3p1pp1/2p2q1p/pp2p3/4P3/1BNP4/PPP2PPP/R2QK1NR w KQkq - 0 8[/fen]

In the above position Igor/Ivan and Super 9 will play 8. Qf3? Which of course is just really really bad since you lose the Bishop and the game before you even started to play the game. The two Horvath's have not shown this move in the above tests.


Other than 8. Qf3? I have not seen any other distinguishing moves to point towards a Spracklen engine with Super 9. The 5 seconds per move should be fairly close to the 1/2 seconds per move in the first test. I might later try to play a higher level than 30 seconds as well with Super 9 to see if there is a closer match.

Best regards
Just for your info Nick
a question arose in another thread about the Mephisto Explorer
as we know the computer is a Morsch program

anyway out of curiosity and since i has the computer already out..i set up this position and guess what..
it plays 8.Qf3 at the 5 second level

I'm Just Sayin Regards
Steve
Hi Nick,
Sorry for my slow reaction, but seeing this position I had to mention that during the rating test the Carnelian II had problems with it too. At 30 seconds it plays 8.Nf3, at 60 secs it plays 8.Qh5, and sticks with that move, even tried it at analysis level. I waited 10 minutes and still Qh5 was the outcome. After 8. .. a4 it plays 9.Bxf7+ Qxf7 10. Qxe5+, which is probably the reason it chose this move. But if black plays 8. .. g6 instead, I guess white gets only one pawn for the bishop.

Carnelian II mysterious ways regards,
Paul
2024 Special thread: viewtopic.php?f=3&t=12741
2024 Special results and standings: https://schaakcomputers.nl/paul_w/Tourn ... 25_06.html
If I am mistaken, it must be caused by a horizon effect...
User avatar
spacious_mind
Senior Member
Posts: 4000
Joined: Wed Aug 01, 2007 10:20 pm
Location: Alabama
Contact:

Post by spacious_mind »

paulwise3 wrote: Hi Nick,
Sorry for my slow reaction, but seeing this position I had to mention that during the rating test the Carnelian II had problems with it too. At 30 seconds it plays 8.Nf3, at 60 secs it plays 8.Qh5, and sticks with that move, even tried it at analysis level. I waited 10 minutes and still Qh5 was the outcome. After 8. .. a4 it plays 9.Bxf7+ Qxf7 10. Qxe5+, which is probably the reason it chose this move. But if black plays 8. .. g6 instead, I guess white gets only one pawn for the bishop.

Carnelian II mysterious ways regards,
Paul
Hi Paul,

Yes there are very many older computers that will have a problem in seeing that the Bishop is trapped unless a3 or a4 is played. The uniqueness that I am alluding to is however 8. Qf3 followed by 9. Qf3. There are many other options that other programs play as you showed where they do Qh5 or Nf3 or other moves.

Best regards
Nick
User avatar
paulwise3
Senior Member
Posts: 1505
Joined: Tue Jan 06, 2015 10:56 am
Location: Eindhoven, Netherlands

Post by paulwise3 »

spacious_mind wrote: Hi Paul,

Yes there are very many older computers that will have a problem in seeing that the Bishop is trapped unless a3 or a4 is played. The uniqueness that I am alluding to is however 8. Qf3 followed by 9. Qf3. There are many other options that other programs play as you showed where they do Qh5 or Nf3 or other moves.

Best regards
Hi Nick,

Ok, I missed your point. What I was trying to say (after playing a number of games against other machines) is that the Carnelian II seems to have a narrow minded selective search that is aimed against the enemy king. Did I already mention that I did the Colditz test with it, and scored an amazing 1755 elo? But in normal games I guess it hardly gets to 1400 elo, also due to horrible endgame play.

Still testing the Carnelian II regards,
Paul
2024 Special thread: viewtopic.php?f=3&t=12741
2024 Special results and standings: https://schaakcomputers.nl/paul_w/Tourn ... 25_06.html
If I am mistaken, it must be caused by a horizon effect...
Post Reply