Author: Don Dailey
Date: 18:15:24 07/20/99
Go up one level in this thread
>>But what really stood out was that I was expecting the difference to >>be trivial, after all I was comparing to a good PVS implementation. I >>really don't remember the exact numbers, but it was substantial and >>way too much to ignore. It was over a 15% speedup. >> >>If I used MTD with NO lazy evaluation versus PVS with lazy evaluation, >>PVS would be a win because lazy evaluation is a big win. If you >>measure improvement by the number of nodes needed to do a given >>iteration then MTD is a big win there is simply no question about >>this, certainly not with Cilkchess. Lazy evaluation is a nodes per >>second optimization and I have no reason to believe it would suddenly >>not work with MTD. MTD is simply another kind of aspiration search. > >I'm not sure first that I'm understanding why you can't use lazy eval with MTD. >Is it because you trash your hash table and get search instabilities? If so, >that sounds like a serious practical problem. > >I'm not sure why it's fair to factor out lazy eval. My own goal here is not to >do the fewest nodes necessary to reach depth D, my goal is to get to depth D in >the least amount of time. And if PVS + lazy eval < MTD, I'll take PVS + lazy >eval. I don't I got my point across very well. I am saying there is no reason whatsoever that you cannot use lazy evaluation with MTD. It was a problem earlier because I did not understand fully what I was doing but I think the problem is solved now. So I'm saying that I don't have a good excuse NOT to use MTD and that you can indeed use lazy evaluation with no problem. - Don
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.