|
|
Post by Steve Draper on Aug 30, 2014 15:11:53 GMT -8
The new Skirmish variant uses GDL based on that of Speed Chess. However, their simulation rates are very different. I get approximately 8000 playouts/second in Skirmish per thread, but only 2000 in Speed Chess. I have not yet looked into what might be causing this, but it seems rather odd since (AFAIK) they don't really differ apart from in goals (and the Speed Chess goals are, if anything, simpler). Anyone got any ideas? Alex?
|
|
|
|
Post by alandau on Aug 30, 2014 17:33:16 GMT -8
My initial testing showed it to be slightly slower on the state machines I tested, and you're correct that the rule changes shouldn't be making them faster.
Are your figures coming from over the course of the game as a whole, as opposed to just similar states of the game? Skirmish will tend to last longer and spend more time in board states that are mostly empty.
|
|
|
|
Post by Steve Draper on Aug 31, 2014 6:16:01 GMT -8
My initial testing showed it to be slightly slower on the state machines I tested, and you're correct that the rule changes shouldn't be making them faster. Are your figures coming from over the course of the game as a whole, as opposed to just similar states of the game? Skirmish will tend to last longer and spend more time in board states that are mostly empty. Depth charges per second from initial state. My stats show the following (all from random playouts from initial during meta gaming): Skirmish:Game length range: [99,99] average 99 Num playouts/second: 8227 Speed Chess:Game length range: [21,199] average 196 Num playouts/second: 2606 Obviously the doubled game length in Speed Chess is a large part of the answer, and hat increased length will mostly be in sparser states. Consequently I agree with the reasoning you proposed, but in favor of Skirmish not of Speed Chess. Not that this is purely from random playouts - directed search will typically lead to shorter games, and may well make the two about equal in practice. |
|
