Schooner 2

[Gabor Szots]

Dennis, you have taken me in. I abandoned my first gauntlet with Schooner but in the second one against average 3050 opponents it still scores 70 %...

So it seems you've made a 600 Elo improvement. That's cruel.

[BrendanJNorman]

Dennis, you have taken me in. I abandoned my first gauntlet with Schooner but in the second one against average 3050 opponents it still scores 70 %...

So it seems you've made a 600 Elo improvement. That's cruel.

And hints as to the CCRL rating? Surely it'll be over 3200!

Impressive is an understatement - especially to do such work and retain such a unique playing style.

[D Sceviour]

Dennis, you have taken me in. I abandoned my first gauntlet with Schooner but in the second one against average 3050 opponents it still scores 70 %...

So it seems you've made a 600 Elo improvement. That's cruel.

As I posted several times before, it is whole new engine. It has been re-programmed from the ground up. It is hard to start with who to thank: Claude Shannon, Bob Hyatt, HG Mueller, the Stockfish team, Andrew Grant, Fabien Letouzey, Milos Tatarevic and countless others. The testers are producing millions of games, and a programmer cannot afford to ignore the results they are finding. And of course, I have added my own special routines. As to the final rating, I have no idea yet. It started losing badly to Texel in Graham Banks tournament. I have been testing the beta version on HG Mueller's ICS monthly tournament, and the rating increase was about 300 elo.

[BrendanJNorman]

Wow Schooner is at least 2800, probably 2900 now!

Isn't that like +300 Elo or more?

Hello Brendan,
Yes, 150+ just for the new gcc v7.2.0-posix compiler (the old one was no good). This was recommended by Andrew Grant. The new compiler required a complete re-write of the program so a lot of other things got fixed.

It has a really nice playing style too, reminiscent of Deep Junior back in the days.

Enjoying watching this engine and even played a game against it.

It was only 3 2, but I switched off PV (so I cant see his eval/lines) and was doing okay until blundering with the ridiculous 33...Ke8 (a better, but similar plan would be to move the king to h8 and prepare ...g5 in that way...where to put the king was my dilemma with seconds on the clock).

Looks kind of like an anti-computer game where the engine is black.

Nice game. Interesting closed formation.

I am not sure where I would like to the style to go in the future. With the deeper selective searching now available, it tends to not find the best move. Instead it selects a move only good enough to win. It misses on some of my favorite position tests. On other ones it perform well. For example in Fine 70, it finds mate in only 3.02 seconds on my machine. That even beats Stockfish 9.

[d]8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - 0 1 bm Kb1 ; Fine 70

Thanks Dennis,

I wasn't deliberately trying to "block up the position" but the moves chosen just seemed best.

Knowing when to play ...exd4 and when to play ...e4 in such structures needs some decent understanding.

Anyway, I played the game before testing Schooner against other engines, so expected a minor update to a 2500 Elo engine - I had no idea that the beast I was playing was now over 3000!

Could you share some of the things you changed besides the compiler? This improvement is amazing to me as a non-programmer.

P.S. There are several types of position that Schooner now seems to "see" the truth before Stockfish - if you keep up this rate of improvement, you're gonna catch him!

[Jim Collins]

I have loaded Schooner into Fritz by using wb2uci. But no matter what time setting I use it makes all moves in 1 sec. Anyway to fix this?

[D Sceviour]

I have loaded Schooner into Fritz by using wb2uci. But no matter what time setting I use it makes all moves in 1 sec. Anyway to fix this?

I do not have Fritz installed, so I cannot help you there. Schooner should directly support UCI without using wb2uci - if you classify it as such. It cannot analyze or perform epd tests, but it should be able to play a timed game: regular, blitz or incremental chess.

[Tibono]

Thanks Dennis for sharing Schooner 2.
I report a strange behavior, setting ponder on in sc.ini (running as Winboard engine, using Arena or Winboard): pondering occurs but appears not to trigger an unlimited analysis (stops after roughly getting to similar time and depth as a move turn); in addition playing the forecasted move triggers a full search restart from depth 1 (maybe speeded up by the hash table fed during pondering?).
Schooner seems never to profit by a ponder hit for immediate answer (saving time for further use).
And depth reached during the thinking time next to a ponder hit does not exceed the one reached while pondering the same move, thus gain seems not obvious!
Warm regards,
Tibono

[Jim Collins]

Thanks Dennis. My mistake on the install.

[D Sceviour]

Thanks Dennis for sharing Schooner 2.
I report a strange behavior, setting ponder on in sc.ini (running as Winboard engine, using Arena or Winboard): pondering occurs but appears not to trigger an unlimited analysis (stops after roughly getting to similar time and depth as a move turn); in addition playing the forecasted move triggers a full search restart from depth 1 (maybe speeded up by the hash table fed during pondering?).
Schooner seems never to profit by a ponder hit for immediate answer (saving time for further use).
And depth reached during the thinking time next to a ponder hit does not exceed the one reached while pondering the same move, thus gain seems not obvious!
Warm regards,
Tibono

Hello Tibono,
I am glad you brought up the topic of pondering. Pondering does not work in UCI mode. At least it is not supposed to work. If you have managed to fool the uci engine to get it to ponder, then it will probably error. Turn off pondering. If you want to use ponder hits, then you should use a different uci engine.

In Winboard, Schooner ponders positions - not moves. That way the hash table gets filled up with potential responses, rather than showing nothing on the opponents different response. For me, this is a stronger and more useful method for ponder.

[Dann Corbit]

Something odd when analyzing. I guess only game playing is tested.
I gave a Winboard and UCI configuration ten seconds to analyze two positions using Arena.

Code: Select all

Analysis of Q:\s4.epd   
2018-11-02 1:17:36 PM Level: 10 Seconds
Analyzing engine: Schooner2.0.33-sseUci

 1                      
    Avoid move: 
    Best move (Schooner2.0.33-sseUci): Nc3-e4
    Not found in: 00:10
   Best move: Nc3-e4
   2018-11-02 1:17:41 PM, Time for this analysis: 00:00:01, Rated time: 00:10

 2                      
    Avoid move: 
    Best move (Schooner2.0.33-sseUci): Qd8-e7
    Not found in: 00:10
     16	00:00	     773,062	14,866,576	+1.60	Qe7 Qxd5 Rf8 Rf1 Rxf7 Qxf7+ Qxf7 Rxf7 Kxf7 Nc3 Bd4 Na4 Bxb2 Nxb2 c6 Rc1
     17	00:00	     941,781	14,715,328	+1.60	Qe7 Qxd5 Rf8 Rf1 Rxf7 Qxf7+ Qxf7 Rxf7 Kxf7 Nc3 Bd4 Na4 Bxb2 Nxb2 c6 Rc1 Ke6
     18	00:00	   1,297,131	13,102,333	+1.60	Qe7 Qxd5 Rf8 Rf1 Rxf7 Qxf7+ Qxf7 Rxf7 Kxf7 Nc3 Bd4 Na4 Bxb2 Nxb2 c6 Rc1 Ke6 Nd3
   2018-11-02 1:17:43 PM, Time for this analysis: 00:00:01, Rated time: 00:20

0 of 2 matching moves
2018-11-02 1:17:44 PM, Total time: 12:00:07 AM
Rated time: 00:20 = 20 Seconds

--------------------------------------------------------------------------------

Analysis of Q:\s4.epd   
2018-11-02 1:17:36 PM Level: 1 Second
Analyzing engine: Schooner2.0.33-sseWb

 1                      
    Avoid move: 
    Best move (Schooner2.0.33-sseWb): Qd8-e7
    Not found in: 00:10
   2018-11-02 1:18:03 PM, Time for this analysis: 00:00:10, Rated time: 00:10

 2                      
    Avoid move: 
    Best move (Schooner2.0.33-sseWb): Qd8-e7
    Not found in: 00:10
   2018-11-02 1:18:17 PM, Time for this analysis: 00:00:10, Rated time: 00:20

0 of 2 matching moves
2018-11-02 1:18:18 PM, Total time: 12:00:41 AM
Rated time: 00:20 = 20 Seconds

GUI -> Engine
Engine -> GUI
conversation is attached.