Author: blass uri
Date: 20:46:38 10/14/99
Go up one level in this thread
On October 14, 1999 at 17:44:08, Robert Hyatt wrote: >On October 14, 1999 at 11:16:15, blass uri wrote: > >>On October 14, 1999 at 10:18:52, Robert Hyatt wrote: >> >>>On October 14, 1999 at 00:02:43, blass uri wrote: >>> >>>>On October 13, 1999 at 21:36:54, Robert Hyatt wrote: >>>> >>>>>On October 13, 1999 at 19:50:41, Dave Gomboc wrote: >>>>> >>>>>>On October 13, 1999 at 12:18:57, Robert Hyatt wrote: >>>>>> >>>>>>>On October 13, 1999 at 03:34:11, Shep wrote: >>>>>>> >>>>>>>>On October 12, 1999 at 13:59:47, Christophe Theron wrote: >>>>>>>> >>>>>>>> >>>>>>>>>>this case 7.5 minutes. In this case I saw instead that Hiarcs, with 1 or 2 >>>>>>>>>>minutes left after move 52, used all its time, with the word TIME appearing on >>>>>>>>>>its clock. I had never seen this before either. In this 8th game it was lost >>>>>>>>>>anyway, but Hiarcs lost on time. >>>>>>>>>> >>>>>>>>>>Enrique >>>>>>>>>> >>>>>>>>>>>Regards, >>>>>>>>>>>Terry >>>>>>>>> >>>>>>>>> >>>>>>>>>Very human-like, this time management! :) >>>>>>>> >>>>>>>>I have seen this often, especially in Hiarcs 6. At move 32, its eval drops from >>>>>>>>+0.1 to -1.5 and it suddenly uses like 30 minutes for that move, so that only 5 >>>>>>>>minutes remain for the last 8 moves. In other words, Hiarcs is the only program >>>>>>>>that really gets into time trouble like a human. >>>>>>>> >>>>>>>>--- >>>>>>>>Shep >>>>>>> >>>>>>> >>>>>>>This still sounds like a _serious_ "bug" to me. -1.5 is not absolutely losing >>>>>>>by any measure. But 5 minutes for 8 moves probably is losing against a good >>>>>>>opponent with way more time. >>>>>>> >>>>>>>A fine line exists between using just enough time to save yourself, and using >>>>>>>too much time and basically leaving the game resignable. >>>>>> >>>>>>A drop from 0.1 to -1.5 is an excellent time to start panicking. >>>>>> >>>>>>Dave >>>>> >>>>> >>>>> >>>>>Yes... but with caution. -1.5 isn't necessarily losing. Time == 0 is >>>>>absolutely losing. >>>> >>>>Time==0 is absolutely losing but 5 minutes for 8 moves at tournament time >>>>control is not close to be absolutely losing. >>>> >>>>If I suppose the opponent has 24 minutes for 8 moves then this case is better >>>>than 24/5 hardware advantage for 8 moves because of the permanent brain and even >>>>24/5 hardware advantage for the all game and not only for 8 moves is not >>>>absolutely losing. >>>> >>>>Uri >>>> >>>>Uri >>> >>> >>>Continue the math. 5:1 time odds is about 200 Elo difference. Which means >>>a 3 of 4 chance of losing. That is nearly absolute. It is _much_ worse than >>>the odds of losing a game from a pawn down. >> >>1)5:1 hardware advantage is about 200 Elo difference but 5:1 time odds is less >>than 5:1 hardware advantage because of the permanent brain. >> >>2)5:1 time odds for only 8 moves is clearly less than 5:1 time odds for all the >>game. >> >>Uri > > >Very unlikely in point 1. You just failed low. The position is therefore >tactical. And you failed low because your last iteration (a deep search) >found something the previous search didn't. It is unlikely you will do >well in predicting in such positions if you only have a part of a minute to >search while your opponent has 3-5 minutes... I know that when hiarcs fails low it often does not stop searching at the end of iteration and does another iteration and this is the reason that it goes to time trouble. I do not know if the decision is right but I can understand this decision because there is a big chance that when you fail low you do not find the right move because of the horizon affect. suppose you see that you fail low and discovers that your opponent can do checkmate and find a way to lose only a pawn. It is possible that this is only an illusion because of the horizon affect and if you do another iteration you discover that the move that is losing a pawn is not going to save you and that there is another move that is going to save you. Uri
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.