< Earlier Kibitzing · PAGE 565 OF 598 ·
Later Kibitzing> |
Jan-16-14
 | | AylerKupp: <<DcGentle> I wanted to demonstrate that engine evals are exaggerating sometimes, and the general rule I have noticed already is: Positional moves, which don't apparently lead to any concrete advantage the engine can see, are evaluated worse.> Sure, engine evals are sometimes exaggerated and sometimes (although I think less seldom) they are understated. And the amount of exaggeration/understatement depends on the engine’s evaluation function, and this is different for each engine. And possibly for the type of position; some engines may exaggerate the evals in open positions and understate them in closed positions, others may do the reverse. It would be nice to know if this is indeed the case and what each engine does, but that hasn’t been done systematically to my knowledge. The way I see it, it’s basically a guessing game at this point in time. And, of course, an engine can only evaluate what it can see; and specifically the position at the end of a line. For positional moves which typically require many plies for their effect to be evident the engines cannot find their advantage or disadvantage unless they search deep enough, and this will likely take a prohibitively long time. The only practical approach I know of is to slide forward a few moves at a time, but this requires judgment as to which lines to select for sliding and how many moves ahead to slide. It is still an art and requires skill to get optimum results. As far as not believing all evals, the solution to that is easy. Believe <no> evals until the line has been reviewed and validated. As a minimum, I will try to restart the analysis at the last move of a line; if the eval of the resulting position is comparable to the initial eval, then confidence in the line can be justified. But often the eval of the resulting position is less than the original eval, in which case the original eval was exaggerated. Or as in the current position (or likely in any won position), the evals should rise, indicating that they were originally understated. I would assume that many positional moves/approaches will fall in this category. |
|
Jan-16-14
 | | AylerKupp: <iatelier> Specifying the MPV level is not a function of the GUI, it is a function of the engine, if it supports it. One way to do change any of the UCI parameters with Arena 3.0 (I have downloaded Arena 3.5 but have not yet had the time to properly configure it since they changed a lot of things) besides the way that <Tiggler> indicated (which, BTW, I didn’t know) is to first load the engine you want and press Ctrl-1 (if you had two engines loaded, as for a tournament, and you wanted to change the UCI parameters of the second engine, you would press Ctrl-2, etc.). This will display the UCI parameters that the engine supports, and they vary from engine to engine. Most of the parameters I leave alone since I have no idea what they do and how changing them will affect the engine’s playing strength, and in an unknown way. But the ones that I commonly change are the MPV, the number of threads/CPUs (which should be set to the number of cores your CPU chip has since some engines don’t do an autodetect and change this automatically), the hash table size (my desktop computers have 4 GB RAM and I set Hash=1024), and set Ponder=FALSE. I also specify the path to the Gaviota tablebases and tablebase has (I typically specify 256 KB) if that’s what the engine uses (Nalimov tablebase paths and hash table size are specified in Arena 3.0). Another way of doing it with Arena 3.0 is to edit the file ArenaENG.cfg located in the directory where Arena.exe is stored with a text processor. This file records all the UCI parameters for each engine that you have installed so that if you change one of them using Ctrl-1, its change will remain in effect the next time you load that engine. If you are making a lot of similar changes (say, changing the Hash size from the typical defaults of 32 or 64 to 1024), this is a more efficient way of doing it. But leave the front of the file alone (the lines after the [Engine] keyword), change only the lines below the keyword for each engine. Of course, if you want to play it safe, make a copy of the file before changing it. Good luck and let me know if you have any other questions by either posting the question at my forum or sending me an email at aylerkupp@gmail.com. |
|
Jan-17-14 | | chesssalamander: What about 33....Rd6? |
|
Jan-17-14 | | DcGentle: <chesssalamander>: Now it's a bad move, <33... Rd6 ?>, because White won't resign but play on for a draw, and justifiably so. For Black will have a hard time to exchange any pieces without any disadvantage, and an ending with 4 rooks might emerge. No thanks. |
|
Jan-17-14 | | chess gunners: How to add diagram there? 33...♗e3 diagram I almost think that black wins |
|
Jan-17-14 | | chess gunners: Hey! i can't use diagram, how it can be use? |
|
Jan-17-14 | | DcGentle: <chess gunners>: Look here: FEN Help Page |
|
Jan-17-14 | | chess gunners: ([bad FEN: 8/5P1BK/5p1P1/2RP2p1/2P2b2/pRp2P1p/1pk2p2/1r2r3])
Now black would win by 33...♗e3 but 33... ♖ge8 not so bad |
|
Jan-17-14 | | chess gunners: [(FEN:1r2r3/1pk2p2/2gP2b2/2RP2p1/5P1BK/8)] |
|
Jan-17-14 | | chess gunners: Huh.... how to write yea?? |
|
Jan-17-14 | | hms123: <chess gunners>
These two sections
/5P1BK/5p1P1
each have 9 positions rather than 8.
e.g., 5+P+1+B+K =9 spots. |
|
Jan-17-14 | | hms123: 3r2r1/2p2kp1/p1P2pRp /5P2/1p2PR2/1P1pb2P/KB1P4/8
(take out the space)
<chessgunners> Also, it looks like you started at the wrong end of the board. |
|
Jan-17-14
 | | AylerKupp: <chess gunners> Here is a link that might help you: http://www.chess-poster.com/english.... Enter the position by clicking first on the piece and then on the square. The FEN string in the box will be updated each time you add, delete, or move a piece. When you're done, copy the FEN string to the Kibitz box. A Spanish version is also available if that is your native language. |
|
Jan-17-14 | | DcGentle: It would be really nice, if the java applet, which displays a game, could also deliver a FEN. But in order to set up a position and get a FEN from it, one can also use this online board: http://www.lokasoft.nl/tbweb.aspx The easiest way is to use a chess program on the own PC in order to paste & copy FEN's into postings. |
|
Jan-17-14 | | DcGentle: This is the FEN of the current position:
3r2r1/2p2kp1/p1P2pRp/2b2P2/1p2PR2/1P1p3P/KB1P4/8 b - - 0 33  click for larger view |
|
Jan-17-14 | | lost in space: <<hms123:> 3r2r1/2p2kp1/p1P2pRp /5P2/1p2PR2/1P1pb2P/KB1P4/8 (take out the space) >
NOOOOO! This is were I am lost |
|
Jan-17-14 | | benjinathan: <chessgunners> another easy way to do a diagram is to click on "review game" (above, below the board). Then make the moves you wish to make by moving the pieces: so I did 33...Be3 34.dxe3 d2 35.Rf1 d1=Q.
Then right click on the board.
Click on "Copy Position"
and voila it is saved and you can right click on your kibbitz, click paste and here it is:  click for larger view |
|
Jan-17-14 | | chesssalamander: <DcGentle> But, don't we pick up that c6 pawn? (After 33....Rd6) |
|
Jan-17-14 | | hms123: <lost in space>
Sorry. What was I thinking? ;-) |
|
Jan-17-14 | | Ceri: <dunamisvpm: In my humble opinion, I think the World team of amateurs (and that includes me) has an unofficial Elo rating of 2600+ Would anyone refute my opinion for the sake of trying to Gauge the strength of Our Team. Thanks and may the FORCE be with us.> It depends upon comparing like with like.
If RV were to run his strongest engine and we played that, the opponent would be notionally over 3,000 and we would often win. However, the cc world has ratings in the 2,500 - 2,700 range for top players and that is the world we are inhabiting. We include players in our number who approach or exceed 2,500 ICCF, I think, so the only downside is the possibility of the voting system causing us to stray from what some of our stronger players see as best play. So, I guess that you are approximately correct.
On the LetsplayChess server szoydd achieved a rating of 3110 when he was at the top of their rankings and I (later) got to 3166. If you subtract about 600 points you would probably arrive at an equivalent cc rating. I played Michael Buss twice (USCF 2428 correspondence). He lost in quick time as Black and more slowly with White. Cheers,
Ceri |
|
Jan-17-14 | | DaringSpeculator: <hms123: <lost in space>
Sorry. What was I thinking? ;-)>
I think you should send him a copy of this book from your library!...:-) http://www.amazon.com/Theres-No-Pla... |
|
Jan-17-14 | | hms123: <DaringSpeculator> heh. |
|
Jan-17-14 | | thegoodanarchist: This surely is the most lopsided vote in the history of this team! |
|
Jan-17-14 | | DcGentle: <chesssalamander: <DcGentle> But, don't we pick up that c6 pawn? (After 33....Rd6)> Can you do this without White getting the d-pawn? Looks unlikely. |
|
Jan-17-14 | | zsoydd: Dear Ceri,
my peak rating 'there' was 3169. At that time I was a beginner of 2400 ICCF ELO at most. Best regards
zsoyd |
|
 |
 |
< Earlier Kibitzing · PAGE 565 OF 598 ·
Later Kibitzing> |