Author: Robert Hyatt
Date: 07:10:47 10/17/00
Go up one level in this thread
On October 17, 2000 at 07:15:13, Enrique Irazoqui wrote: >On October 17, 2000 at 07:08:32, Uri Blass wrote: > >>On October 17, 2000 at 05:56:11, Enrique Irazoqui wrote: >> >>>On October 16, 2000 at 21:12:58, Robert Hyatt wrote: >>> >>>>On October 16, 2000 at 17:06:38, Enrique Irazoqui wrote: >>>> >>>>>On October 16, 2000 at 15:38:56, Robert Hyatt wrote: >>>>> >>>>>>On October 16, 2000 at 14:05:52, Chessfun wrote: >>>>>> >>>>>>>On October 16, 2000 at 00:53:06, Robert Hyatt wrote: >>>>>>> >>>>>>>>My take: let's wait until the thing is released and see how it does. Without >>>>>>>>beta testers that exert a bit of influence over the program's time allocation >>>>>>>>and book choices. >>>>>>> >>>>>>> >>>>>>>Can you prove this statement please; >>>>>>> >>>>>>>There are no beta testers who exert any influence over book choice or >>>>>>>time allocation. The program runs automatically on the server, the book >>>>>>>is set. >>>>>>> >>>>>>>Sarah. >>>>>> >>>>>> >>>>>>I base it on the following. I have played multiple gambit tiger clones. They >>>>>>are reasonably predictable in their time usage. With a "couple" of beta >>>>>>testers, the thing will go into a "deep think" that is _far_ longer than the >>>>>>time one would expect for a move. IE it moves consistently at 30-50 seconds per >>>>>>move, then takes 10 minutes. In a position where it did _not_ fail low. I >>>>>>believe that the operator simply wants to give it a chance to find something >>>>>>that may (or may not) be there. >>>>>> >>>>>>It is my opinion, with no proof of course. But it is _very_ common with _all_ >>>>>>engines. ChessPartner makes it trivial for the operator to influence things. >>>>>>I can do it with xboard if I thought that I somehow might know more about when >>>>>>to spent more time than Crafty does... >>>>>> >>>>>>As far as proof, simply play a few fully automatic games at (say) 30 30 or >>>>>>whatever time control you like, and then check the times. See if you see any >>>>>>case where it takes more than 10x the normal time per move, when the score did >>>>>>_not_ drop _or_ rise during that search. If you find such cases, I will >>>>>>certainly retract my statement. But in watching so many games of late, it >>>>>>is obvious that something goes on "from time to time". IE I see most programs >>>>>>taking 2x-3x on fail lows. And sometimes for other reasons. But not 10x or >>>>>>longer. >>>>> >>>>>I have seen it often enough, mostly under panic, but not always. Assuming that >>>>>auto232 and automatic FICS are similar, you will see these 10x and bigger from >>>>>time to time. The longest I have seen was almost 16x, but I don't remember if it >>>>>was in panic time. >>>>> >>>>>When starting an auto232 match in DOS, it is good to set the /t parameter >>>>>(timeout) to at least 10x, or else too many games will be terminated before they >>>>>should. SSDF people can confirm. >>>>> >>>>>Enrique >>>>> >>>> >>>>I don't believe this was a panic. I let crafty search it overnight and the >>>>score remained constant. IE there was no reasonable explanation for taking >>>>such a long time... except that the position looked "right" for a deep >>>>combination, one just wasn't there... >>>> >>>>Have you seen that sort of 'deep think'? >>> >>>What a perfect timing that just last night I saw this game Gambit 1.0 - Deep >>>Fritz beta. Time control was 40/20 (30''/move average). Look at Gambit's move >>>29.Qg5+ played after 545 seconds, which is 18x. Gambit wasn't failing at this >>>moment, but had been in book for the first 27 moves and had plenty of time to >>>spare. It doesn't happen often, of course, but now and then you do get these >>>>10x. >> >>The average time per move is 30 second per move but if you consider the fact >>that white was in book for 27 moves then is it 20 minutes/13 moves and the >>average time for these 13 moves is more than 1.5 minute per move(assuming that >>the program use the 20 minutes or almost all of the 20 minutes for 40 moves and >>most programs do it) so it is only 6x and not 18x. > >Still 18x, no matter how you look at it in the context of this thread. Bob was >surprised to see Gambit taking more than 2 or 3x and he doesn't know when it >leaves book. Also, had you been autoplaying this game with a timeout <10x, it >would have been terminated at move 29. 18x it is. > >Enrique > >>Uri That certainly lowers my confidence in my original statement about human intervention in time management... although 18x sounds very dangerous... it _could_ be related to the depth of the book line, which would then change my mind a second time, as the book line I watched in the game I mentioned was only 10 moves, and the deep think came 20 moves later.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.