Everyone on mersenneforum vs. Stockfish - Page 7

LaurV · 2015-11-12, 12:02

Quote:

Originally Posted by Brian-E

I wasn't particularly talking about openings, and I didn't think LaurV was particularly, either. All I mean is that chess programs won't always select the same move from a given random position. So playing the same game against Stockfish again (to see if it blunders again) will necessitate lots of corrections to keep the game on course.

EDIT: I guess you're making the point that even if programs were predictable, the match games would be varied by the different openings. Agreed.

Yes exactly! (on both accounts, i.e. including the edit).

LaurV · 2015-11-12, 12:10

Quote:

Originally Posted by axn

I am not sure whether there is any randomness actually built into SF

Yes it is. The source code is open, so you can look if you like, and no, I didn't look, so I may be wrong. But in a random position where there are more "best moves" within close range of ponders, all engines I know will select a random move from those "at par". For some engines you can set this randomness, and what "at par" means (i.e. how much different should be the score allowed to vary, to be considered "the same value" of the move).

Else, as Brian/retina said already, you beat the program once, eventually taking moves back, then you write it down and you can beat it every time playing exactly the same game. This is not happening with engines developed in the last 30 years (it was a frequent symptom for very old chess engines, on Z80 and XT computers, as I remember).

MooMoo2 · 2015-11-13, 05:32

Quote:

Originally Posted by Brian-E

Here's a vaguely encouraging article for those of us who feel our chances of avoiding defeat are near zero.

http://www.chess.com/blog/smurfo/com...t-blunderright

Stockfish has been (and will be) using only one thread during the entire game, so it shouldn't be affected by the bug:

Quote:

Ensure that rootDepth < DEPTH_MAX
Indeed, if we use a depth >= DEPTH_MAX, we start having negative depth in the TT (due to int8_t cast). No functional change in single thread mode

I originally decided to use one thread because the default was set to one thread, not because I knew about the bug. Also, using one thread allowed me to open up and run multiple instances of Stockfish at the same time, each assuming a different move by white. That way, I can post Stockfish's response immediately after the polls close.

LaurV · 2015-11-13, 14:47

Re: K vs SF game, randomness, replaying it, etc.
We succeeded to dld the stockfish and attach an interface to it. We hope this will not impact the future game (the temptation is huge, you know...)
Anyhow, we tried to replay the game 22, as per Brian's link and the further link in the article linked by him.
First, you MUST click that link.
You will need a huge monitor, the lower-side of the screen graphics do not fit on our monitor. But we can scroll.
Looking to those graphics, you see the "depth" searched by each engine, in half-moves.
Now, my depth, even if all 4 cores are used, are only 1/3 of the depth in that graphic, after hour(s) of "thinking".
So you see, no way to play "the same game".

After placing the first 8 moves on the table, as indicated by the game rules, just from move 9 (which technically is the first white move) SF refuses to do Bg2. He indicates a series of different moves, depending on the thinking depth, from which the predominant is 9. Bd3. Even if Bg2 is in all the books, it appears as "the best" only when SF reaches a depth of ~22, which takes more than an hour. We had to switch from a wide tree to a narrow tree, to speed up. That is, we played with engine's settings, a part of which we have no idea what they are for.

Interesting is that after a 3-4 moves start, SP plays the opening right, up to the move 9, so it would make no difference for this particular game if we start with 3 moves played or with 8 moves played like the rules currently are.

There are another 12-15 moves during the game that had to be "forced by hand", because "my" SF was indicating a totally different line of thinking (most probably the cause is the depth, we never reached behind 23, and we only saw 25+ when the game was close to the end. But the competition "depth" was mostly over 35, and occasionally close to 50 semi-moves, if we believe those graphics).

Other interesting point, according with the depth we played, white was in the negatives (like -0.8 and -1.2 PV!) somewhere in the middle of the game after we forced one of the moves to be according with the played game. That is not a joke! Only going deeper would (most probably) prove the fact that such moves were actually the best. This is somehow fascinating, when you think about...

Reaching the incriminated move, our SF wanted to move 64. Bc2 (!) from the start, it never tried other moves. Here we believed that after 64...Qxc6 the game is a draw. Indeed, after 65. Qxe5+, Kg8 66. Qxb2, in spite of the fact that black just lost the rook, and white has two pawns and a bishop in advance, however white lost the "sente" and his king is alone in the lower right side of the board, which allows black to drag it into an eternal check. But to our surprise, SF didn't play 66. Qxb2, but he left the rook alone and played 66 Qg3+, which indeed results in white win after a very complicate itinerary of the black king on the board.

So, it turns out that the game is white win, after all, if 64. Bc2 is played, as said by other people already.

I think that extremely few human players would be able to win this game as white, after move 63. I am not one of those players... if I would be white in that situation in a real match, I would be extremely frustrated that I can't win, contrary to huge material advantage.

A nice puzzle to propose... White plays and wins... How?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Stockfish-assisted game vs Stockfish, move 8 discussion	MooMoo2	Other Chess Games	11	2016-10-21 15:47
ftp.mersenneforum.org	Mark Rose	Forum Feedback	11	2015-05-14 01:44
The mersenneforum First 300	ewmayer	Lounge	15	2012-08-27 18:35
HELP!! Can't Log In to mersenneforum	Unregistered	Information & Answers	8	2010-09-06 18:24
archives of mersenneforum.org?	masser	Forum Feedback	2	2007-12-27 15:34

2015-11-13, 14:47	#70
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 41·251 Posts	Re: K vs SF game, randomness, replaying it, etc. We succeeded to dld the stockfish and attach an interface to it. We hope this will not impact the future game (the temptation is huge, you know...) Anyhow, we tried to replay the game 22, as per Brian's link and the further link in the article linked by him. First, you MUST click that link. You will need a huge monitor, the lower-side of the screen graphics do not fit on our monitor. But we can scroll. Looking to those graphics, you see the "depth" searched by each engine, in half-moves. Now, my depth, even if all 4 cores are used, are only 1/3 of the depth in that graphic, after hour(s) of "thinking". So you see, no way to play "the same game". After placing the first 8 moves on the table, as indicated by the game rules, just from move 9 (which technically is the first white move) SF refuses to do Bg2. He indicates a series of different moves, depending on the thinking depth, from which the predominant is 9. Bd3. Even if Bg2 is in all the books, it appears as "the best" only when SF reaches a depth of ~22, which takes more than an hour. We had to switch from a wide tree to a narrow tree, to speed up. That is, we played with engine's settings, a part of which we have no idea what they are for. Interesting is that after a 3-4 moves start, SP plays the opening right, up to the move 9, so it would make no difference for this particular game if we start with 3 moves played or with 8 moves played like the rules currently are. There are another 12-15 moves during the game that had to be "forced by hand", because "my" SF was indicating a totally different line of thinking (most probably the cause is the depth, we never reached behind 23, and we only saw 25+ when the game was close to the end. But the competition "depth" was mostly over 35, and occasionally close to 50 semi-moves, if we believe those graphics). Other interesting point, according with the depth we played, white was in the negatives (like -0.8 and -1.2 PV!) somewhere in the middle of the game after we forced one of the moves to be according with the played game. That is not a joke! Only going deeper would (most probably) prove the fact that such moves were actually the best. This is somehow fascinating, when you think about... Reaching the incriminated move, our SF wanted to move 64. Bc2 (!) from the start, it never tried other moves. Here we believed that after 64...Qxc6 the game is a draw. Indeed, after 65. Qxe5+, Kg8 66. Qxb2, in spite of the fact that black just lost the rook, and white has two pawns and a bishop in advance, however white lost the "sente" and his king is alone in the lower right side of the board, which allows black to drag it into an eternal check. But to our surprise, SF didn't play 66. Qxb2, but he left the rook alone and played 66 Qg3+, which indeed results in white win after a very complicate itinerary of the black king on the board. So, it turns out that the game is white win, after all, if 64. Bc2 is played, as said by other people already. I think that extremely few human players would be able to win this game as white, after move 63. I am not one of those players... if I would be white in that situation in a real match, I would be extremely frustrated that I can't win, contrary to huge material advantage. A nice puzzle to propose... White plays and wins... How?