Solving chess one position at a time

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Mario Moran-Venegas
    replied
    Originally posted by Roger Patterson View Post

    You might be interested in a recent Perpetual Chess interview with Matthew Sadler. He discusses the current state of this idea, not in the context of correspondence chess, but in the context of world championship preparation.

    He also has some interesting things to say about Leela playing without doing any search (i.e. playing purely on an evaluation of the position after 1 ply). Apparently Leela is quite strong (better than me anyway) with only a 1 ply search.

    https://www.perpetualchesspod.com/ne...matthew-sadler
    Not entirely correct . Leela or any neural network is worthless unless trained. When leela plays a chess game it is misleading to say is at 1 ply search. The file containing the chess results is ready when 1 billion games or so have been played by it against itself.

    Leave a comment:


  • Roger Patterson
    replied
    Originally posted by Egidijus Zeromskis View Post



    Does a human interference help in this process these days?
    You might be interested in a recent Perpetual Chess interview with Matthew Sadler. He discusses the current state of this idea, not in the context of correspondence chess, but in the context of world championship preparation.

    He also has some interesting things to say about Leela playing without doing any search (i.e. playing purely on an evaluation of the position after 1 ply). Apparently Leela is quite strong (better than me anyway) with only a 1 ply search.

    https://www.perpetualchesspod.com/ne...matthew-sadler

    Leave a comment:


  • Mario Moran-Venegas
    replied
    New preliminary findings for the game:
    1) The black pieces have an advantage.... in the one thousandth of a pawn.(these result in a draw )

    2) There are thousands of unavoidable draws .If you can memorize the sequence you can draw every game.
    Computer-Computer tournaments where they are allowed to use opening BOOKS are finding this out already.

    Leave a comment:


  • Sam Sharpe
    replied
    Deep = ply; how long it takes it reach any given ply depends on the CPU and the position.
    Human involvement doesn't help the evaluation, but with several moves at the same eval, having the engine run against these sub-variations (multiple iterations) sometimes helps in trying to figure out which move to make.

    Leave a comment:


  • Egidijus Zeromskis
    replied
    Originally posted by Sam Sharpe View Post
    not letting engine go deep enough
    Do you mean keeping an engine for a long time?

    Does a human interference help in this process these days?

    Leave a comment:


  • Sam Sharpe
    replied
    Originally posted by Mario Moran-Venegas View Post
    ...
    Sam is the ICCF database freely available? Because you and Pino Verde plus all those who represent Canada in the Correspondence Chess world use engines (as everybody is allowed to do now) for your moves then your moves are gold.
    You can get the ICCF DB games here: https://www.iccf.com/message?message=454

    You will probably want to filter out a few things to get quality games (even if everyone was using the "latest and greatest" engine version):
    - dates ... with the advance in engines, especially when NNUE came in, anything more than a few years old is produced by weaker engines (or people! egads... what a concept!!)
    - ratings ... unless someone is just starting out, if they are low or moderately rated, there is probably a reason (eg. old software, not letting engine go deep enough, etc...)

    Leave a comment:


  • Pargat Perrer
    replied
    Originally posted by Mario Moran-Venegas View Post
    If anybody is interested in an evaluated position exchange pm me here or at latin chess master at g mail dot com.
    Your computer evaluations must go to 60 ply at least. Terms of agreement are flexible.
    Of course if you need evaluations and do not have a computer I can provide them (for a fee ).

    I'm wondering if anyone has done this already: set up the standard chess opening position, then switch Black's King and Queen. Now each side's King is on the same file as the other side's Queen.

    Analyze this using some of these computers you are talking about Mario.

    What I am wondering is if the resulting opening tree would be as extensive as it is for standard chess, or would this setup somehow give too much advantage to one side or the other.

    I started doing this with my 8-core computer and Stockfish 13. I was getting up to 36 plies deep, and the score was something like +0.5 for White, and so I played out that main line and then started the analysis again. Suddenly the score went to -3 point something! I haven't yet narrowed down just where the engine suddenly saw a drastic turn in favor of Black.

    Leave a comment:


  • Mario Moran-Venegas
    replied
    Originally posted by Bob Armstrong View Post
    Mario - 64 is such a significant number!! Have a great one!

    That train left the station for me eons ago!

    Bob A

    P.S. Thanks Frances........
    Bob!! 64 is significant for those who play that easy game. I am a Go player remember? 361 is my number!

    Leave a comment:


  • Mario Moran-Venegas
    replied
    Thank you for the birthday wishes.

    Cloud computing is ok for things you do not care to keep secret. As mentioned above, companies see the value of a deeply evaluated chess position. Intel is running a freebee on cloud computing. But in the fine print... can they copy the results?
    Francis ,you have computers working on music, would you trust running that in a cloud computer?

    My machines are working 25 hours per day.

    Sam is the ICCF database freely available? Because you and Pino Verde plus all those who represent Canada in the Correspondence Chess world use engines (as everybody is allowed to do now) for your moves then your moves are gold.

    Leave a comment:


  • Bob Armstrong
    replied
    Mario - 64 is such a significant number!! Have a great one!

    That train left the station for me eons ago!

    Bob A

    P.S. Thanks Frances........

    Leave a comment:


  • Francis Rodriguez
    replied
    Originally posted by Bob Armstrong View Post
    Mario & Sam - do you guys know at all how much you freak us patzers out??

    Bob A
    Bob, it's your ex-roomie Mario MV's birthday today!
    Mebbe gift him 64 cores (not Apple).....and he turns 64!

    Leave a comment:


  • Sam Sharpe
    replied
    If I could figure that out, I'd have my GM title. LOL

    Leave a comment:


  • Pargat Perrer
    replied
    Originally posted by Sam Sharpe View Post
    Time to reach any given ply depends on the position, Mario. I use a Ryzen 7 3700, 8 core (16 logical), 64 GB cache and a SSD. Cost around 2grand and I generally go 65-70 ply for my ICCF games. Sometimes it takes 3-4 hours to get there, sometimes 18+.
    How do you decide when to stop the analysis and make a move in these ICCF games? Especially for the most complicated positions.

    Leave a comment:


  • Sam Sharpe
    replied
    $3 for 4 hrs = $18/day = $6,570/ per year .... Gadzooks! I run basically 7/24 with CPU @ 100%.

    Leave a comment:


  • Don Parakin
    replied
    Originally posted by Sam Sharpe View Post
    Time to reach any given ply depends on the position, Mario. I use a Ryzen 7 3700, 8 core (16 logical), 64 GB cache and a SSD. Cost around 2grand and I generally go 65-70 ply for my ICCF games. Sometimes it takes 3-4 hours to get there, sometimes 18+.
    Cloud computing is great for stuff like this. You can create a virtual machine, run it for a few hours, then delete it and only pay for the hours it existed. For US$0.57 per hour on Google Cloud Platform you can get a cpu-optimized VM with 16 cores and 14G of memory. So, under C$3 for 4 hours of analysis. No need to have a $10K or $50K real machine sitting idle 99.99% of the time.

    Leave a comment:

Working...
X