mmt Vs. Ovyron (G4 D5 BG2)

Discussion of computer chess matches and engine tournaments.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
jp
Posts: 1331
Joined: Mon Apr 23, 2018 5:54 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by jp » Wed Feb 19, 2020 12:00 am

To be frank, I thought that was a ridiculous dummy spit by you.

You act like it's so personal. The games only have any interest to the extent that they tell us something about 1.g4. It's not about egos.

User avatar
Ovyron
Posts: 3998
Joined: Tue Jul 03, 2007 2:30 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by Ovyron » Wed Feb 19, 2020 12:08 am

jp, you act as if I owe you something. I owe you nothing. I could continue the mmt game in private and you'd not get to see it. I could send my solution to the puzzle to Zenmastur and request that he keeps it private and you don't see it.

This entertainment you're getting is being given for free as a community service, so stop with any demand or the whole thing can be just shut down. I'll solve the puzzle whenever I want and move in this game whenever I want, my opponent or challenger are the only ones with the right of hurrying me up, but as part of the audience the least I can expect from you is appreciation.

jp
Posts: 1331
Joined: Mon Apr 23, 2018 5:54 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by jp » Wed Feb 19, 2020 12:10 am

I'll stop posting about that, because this will go round in circles, delaying things further.

I don't think anyone on this forum "owes me something", apart from basic courtesy and respect that everyone is owed. You really take everything so personally. You'd probably be happier if you didn't.

User avatar
Ovyron
Posts: 3998
Joined: Tue Jul 03, 2007 2:30 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by Ovyron » Wed Feb 19, 2020 12:14 am

jp wrote:
Wed Feb 19, 2020 12:10 am
delaying things further.
Not really, I can interact with the positions while making posts, no delay of me solving the puzzle or playing a move in this game has happened because of my posting on the forum.

It's called "multi-tasking". You should try it. You can do things like posting on a forum while analyzing chess positions, or eating popcorn while watching a movie, think of the possibilities!

Spliffjiffer
Posts: 244
Joined: Thu Aug 02, 2012 5:48 pm
Location: Germany

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by Spliffjiffer » Wed Feb 19, 2020 2:45 am

i wish mmt will be focused, nothing is over nor lost, both sides a happy game...may the better succeed...id like to see a separat thread for discussions beyoned the game, hf and gl
Wahrheiten sind Illusionen von denen wir aber vergessen haben dass sie welche sind.

Uri Blass
Posts: 8756
Joined: Wed Mar 08, 2006 11:37 pm
Location: Tel-Aviv Israel

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by Uri Blass » Wed Feb 19, 2020 11:08 pm

Ovyron wrote:
Tue Feb 18, 2020 11:58 pm
zullil wrote:
Tue Feb 18, 2020 11:51 pm
Maybe just ask politely at https://groups.google.com/forum/?nomobi ... ishcooking ?
When people produce games to test the engine they automatically test for mate claims. You don't need to run any games, just look at mate claims by Stockfish in the PGN and see if at any point it has reported a mate score that was wrong (the defending side can delay mate.)

What jp is talking about has never been witnessed, I don't know from where this fear of a wrongly reported mate comes from.
The fear comes from what we call hash collision.

It is known that stockfish does not store all the position in the hash and it is known that in the past
some patch for stockfish that increased the probability for hash collision passed.

This is the reason that I am afraid it may consider 2 different positions to be the same and memorize that some position is mate in 5 from previous search when practically the mate score is wrong because stockfish never visited that position earlier and visited a different position.

I know of no case that stockfish reported a wrong mate score but I know of no try to test it for example by letting stockfish to search 1,000,000 random tablebases positions without tablebases and telling me that it has 0 wrong mate scores in all the searches(in tablebases positions it is easy to say yes or no for reporting a wrong mate score).

Zenmastur
Posts: 829
Joined: Sat May 31, 2014 6:28 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by Zenmastur » Thu Feb 20, 2020 12:19 am

Uri Blass wrote:
Wed Feb 19, 2020 11:08 pm
Ovyron wrote:
Tue Feb 18, 2020 11:58 pm
zullil wrote:
Tue Feb 18, 2020 11:51 pm
Maybe just ask politely at https://groups.google.com/forum/?nomobi ... ishcooking ?
When people produce games to test the engine they automatically test for mate claims. You don't need to run any games, just look at mate claims by Stockfish in the PGN and see if at any point it has reported a mate score that was wrong (the defending side can delay mate.)

What jp is talking about has never been witnessed, I don't know from where this fear of a wrongly reported mate comes from.
The fear comes from what we call hash collision.

It is known that stockfish does not store all the position in the hash and it is known that in the past
some patch for stockfish that increased the probability for hash collision passed.

This is the reason that I am afraid it may consider 2 different positions to be the same and memorize that some position is mate in 5 from previous search when practically the mate score is wrong because stockfish never visited that position earlier and visited a different position.

I know of no case that stockfish reported a wrong mate score but I know of no try to test it for example by letting stockfish to search 1,000,000 random tablebases positions without tablebases and telling me that it has 0 wrong mate scores in all the searches(in tablebases positions it is easy to say yes or no for reporting a wrong mate score).
I see this all the time. SF used to store just 16 bits of the hash in the TT. If you are searching 64Mnps and the TT is full you will get about 1000 hash collisions per second. Many of these will get overwritten without being used, but the longer the search goes on the more of these "false" entries will accumulate in the TT. During a mate search the number of transpositions can be absolutely huge. So, these "false" entries will be used by the search and be propagated to "new" entries in the TT. Some of these will be overwritten, just like all the other entries, but since the hash is "random" they are statistically no more likely to be overwritten than the percentage of the TT they occupy.

There are several ways in which a “false” TT entry can be evicted from the TT. It can be replaced by an unrelated entry. The search can need the entry but at a deeper search depth. The move stored in the entry can be illegal. I'm not sure what stockfish does in this case. Does anyone know if SF even checks for illegal moves in the TT and if it does, what it does about it? Hopefully it clears the entry so that it can't be used and thus propagate to entries higher in the search tree. I guess using an illegal move counter for the TT would be one way to keep some sort of track of how many bad entries are found in the TT.

The last time I checked, and it's been a while, SF used 6 10-byte TT entries per 64-byte cache-line which were split into 2 3-entry buckets. This left 4 bytes unused per cache-line. I'm not sure this is still the case because I know the TT has been altered to handle non-power of 2 cache buckets. But, assuming it is true, 4 additional bits for each entry could be allocated to additional hash bits to either, limit the number of hash collisions or, keep better account on the number of hash collisions.

Regards,

Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

Zenmastur
Posts: 829
Joined: Sat May 31, 2014 6:28 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by Zenmastur » Thu Feb 20, 2020 12:27 am

zullil wrote:
Tue Feb 18, 2020 12:43 pm
zullil wrote:
Tue Feb 18, 2020 1:12 am
Zenmastur wrote:
Tue Feb 18, 2020 12:41 am
Ovyron wrote:
Mon Feb 17, 2020 9:34 pm
Now I feel like a cheater :P
I guess I should have put a time limit on it, but at the time I didn't envision it taking nearly this long. I wouldn't worry about cheating too much. Unless they start giving you moves I'm good with it. Zullil's mate is probably wrong anyway. :D :D :D
...
Regards,

Zenmastur
As I said, I did nothing. If mate-in-29 is wrong then Stockfish is to blame. :wink:

Mate-in-28 now, by the way...
Stockfish-dev has held at mate-in-28 for eleven hours now. Depth 100. Will be curious to learn what others have gotten by carefully "guiding" an engine.
Carefully "guiding" and engine during a mate search isn't going to gain you much. There is no way "guiding" an engine can change the length of mate the position contains. The best you can hope for is to speed the search up. The truth be told even speeding the search up is going to be difficult for a human. This is why Ovyron hates this test so much. It's almost purely a hardware test. If you have big hardware and lots of memory you will spend less time on problems like this.
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

jp
Posts: 1331
Joined: Mon Apr 23, 2018 5:54 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by jp » Thu Feb 20, 2020 2:22 am

Ovyron wrote:
Tue Feb 18, 2020 11:58 pm
When people produce games to test the engine they automatically test for mate claims. You don't need to run any games
I doubt any testing is done. It'd take a lot of effort and since it doesn't directly affect the engine's elo, people don't care. (Which engines make mate announcements and which don't?)

Such testing would be only a basic check of course, and not necessarily the most challenging one. People may have impressive hardware, but most of them like running very short TC games, so mate announcements may give the engine a huge margin for error anyway. (e.g. It blitzes out a move in 5 seconds, announcing mate in 18; if it had longer to calculate, it'd have announced mate in 15; the shortest mate is really mate in 12; it has a margin of error of 6 moves; is it really likely in those conditions it's going to make a wrong announcement of mate in 11 for unmotivated humans to try to discover?)

Zenmastur
Posts: 829
Joined: Sat May 31, 2014 6:28 am

Re: mmt Vs. Ovyron (G4 D5 BG2)

Post by Zenmastur » Thu Feb 20, 2020 5:23 am

jp wrote:
Thu Feb 20, 2020 2:22 am
Ovyron wrote:
Tue Feb 18, 2020 11:58 pm
When people produce games to test the engine they automatically test for mate claims. You don't need to run any games
I doubt any testing is done. It'd take a lot of effort and since it doesn't directly affect the engine's elo, people don't care. (Which engines make mate announcements and which don't?)

Such testing would be only a basic check of course, and not necessarily the most challenging one. People may have impressive hardware, but most of them like running very short TC games, so mate announcements may give the engine a huge margin for error anyway. (e.g. It blitzes out a move in 5 seconds, announcing mate in 18; if it had longer to calculate, it'd have announced mate in 15; the shortest mate is really mate in 12; it has a margin of error of 6 moves; is it really likely in those conditions it's going to make a wrong announcement of mate in 11 for unmotivated humans to try to discover?)
Your statements aren't justified.
zullil wrote:
Sun Feb 16, 2020 3:26 pm
Ovyron wrote:
Sun Feb 16, 2020 3:05 pm
zullil wrote:
Sun Feb 16, 2020 12:32 pm
Since you were only required to get within five moves, that should have taken maybe eight hours.
Maybe I got there in 2 hours. But how would I know? I couldn't live with myself if my line was within 6 moves only AND THEN I have to play 1.g4 myself :shock: - so I spend most of my time nowadays "making sure".

Because, what if that position is in reality a mate in 25?? Then I have to do this until at least getting a mate in 30, and I'm still far away from that.

I checked a position that I had solved to a mate in 13 with Stockfish, and after reaching depth 70 it was still just showing a mate in 23. Depth 70 and it's still 10 moves off. So I can't trust it, I have to check these lines myself, and the thing is, you never know when you're done, because there may be some move that mates faster, and in some of these variations, much, much faster (I think the worst I've seen is Stockfish jumping from Depth 50 "mate in 18" to "mate in 8" after I make the improvement; again it's just 10 moves, but I'm starting to doubt even Zenmastur has the fastest mating line. Some of the moves white can play are just insane and mate faster for no humanly discernible reason, like, white gives away a pawn and a knight for nothing but BAM mating 10 moves faster :shock: )
The very reliable fastgm has mate-in-30. If you're within 5 of that, I suggest you return to the game with mmt. Your "fans" are growing restless.

BTW, have you updated Stockfish to include this patch?
Did you goto the link Zullil posted and read it? If not you should. If no one really cared why would they go to the trouble of testing SF against a large set of known mating positions?

They obviously do care, even if it hurts ELO. They want the code to be correct even at the cost of performance. No one wants a bunch of bugs in their code! It's bad programming practice and leads to future problems not being able to be solved because the bugs begin to interact with each other and it makes it impossible to diagnose what the problem is. Therefore, if it's faulty code it gets “fixed” so when the next bug is discovered it is easier to diagnose and remedy.

I think you'll find most programmers agree that correctness of code is FAR more important than a few ELO.

Regards,

Zenmastur
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.

Post Reply