| Author |
Message |
Greg Schmidt (Gschmidt2)
New member
Username: Gschmidt2
Post Number: 33
Registered: 1-2007
| | Posted on Sunday, July 15, 2007 - 9:41 pm: | |
I don’t understand the ZoG rationale of offering both a ‘Thinking Time Per Move’ slider and a ‘Strength’ slider. To me, it seems as though thinking time alone is sufficient for specifying the level of play.
Setting strength to a high level means that the search becomes limited by time anyway.
If the strength is set low, then the engine will complete the search prior to completing its allotted thinking time. If for example a search with strength 2 takes roughly one second, why not simply set the thinking time to one second? Is there really any significant difference that anyone cares about?
My impression of the ‘Strength’ control is that it is unnecessary and confusing. Most chess engines are time based. I am unaware of any other game program that allows the player to specify both.
Do others agree with what I am saying or am I missing some important point?
-- Greg |
Keith Carter (Keithc)
New member
Username: Keithc
Post Number: 141
Registered: 8-2000
| | Posted on Sunday, July 15, 2007 - 10:35 pm: | |
There is a use for the strength slider though probably it does not come up in general play. Each notch on the strength scale represents one ply of look ahead. If thinking time is set to infinite and variance to small then it becomes an exhaustive search. I use these settings to test one manipulated AI against another.
My kids have used a reduced strength setting to get a consistant opponent. They look ahead about 3 ply in Chess. By setting the strength to the third notch the AI looks only as far ahead as they do. A maximum strength setting at one second starts Chess out at 7 ply on their computer which completely out strips them from the get go. |
Greg Schmidt (Gschmidt2)
New member
Username: Gschmidt2
Post Number: 34
Registered: 1-2007
| | Posted on Monday, July 16, 2007 - 12:22 am: | |
I can see an argument for repeatability. Set the variance to 0, kick the time setting up, and search to a fixed ply that won't exceed the time limit. The outcome should be repeatable. Useful when replaying the same game, debugging a move, etc.
I hadn't considered that someone might want to strictly match their own lookahead to the engine. Interesting point.
After experimenting with Chess, I see that at the lower end of the strength setting, there is a finer degree of control that can be achieved using 'strength' as opposed to using the time control which is limited to 1 second time intervals. For higher settings though, if one sets strength to 8 and it takes roughly 1 second, then conversely, setting the time to 1 second results in roughly an 8 ply search. That general relationship probably accounts for most of my bias. Also, I'm probably taking too narrow a view of this behavior and need to remove my implementor glasses for awhile.
Thanks for your thoughts.
-- Greg |
Michael Nolan (Mrwcf)
New member
Username: Mrwcf
Post Number: 1
Registered: 9-2007
| | Posted on Thursday, September 06, 2007 - 11:15 am: | |
Another use of the Strength Slider is to compensate for the lack of any Thinking Time Per Move setting between 3 Minutes and Forever.
When testing new games for playability, especially ones with unequal armies, I often open multiple occurrences of a game simultaneously, and let the AIs play against themselves over long periods of time.
In order to get a thinking time that is greater than three minutes, I set Thinking Time on Forever, and Strength on one or more notches below Expert. Depending upon the game, this results in thinking times that are greater than three minutes, but not infinite. |
|