Suggesting adjusting searching time according to the board state

[111116]

It's usually not necessary to search for too much time on most situation. I guess it would be better that the program spend more time on harder situations. So I suggest adding another simple function to decide how much the time limit should be. (e.g. according to the number of different tiles)

[wjaskowski]

Nice idea. Possibly it would be even possible to learn such a function instead of hand-crafting it.

…

-- Wojciech Jaśkowski, Ph.D. Laboratory of Intelligent Decision Support Systems Institute of Computing Science Poznan University of Technology, Poland http://www.cs.put.poznan.pl/wjaskowski

On 30 June 2017 at 16:05, 111116 ***@***.***> wrote: It's usually not necessary to search for too much time on most situation. I guess it would be better that the program spend more time on harder situations. So I suggest adding another simple function to decide how much the time limit should be. (e.g. according to the number of different tiles) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEuSHRNFkFv2tZ_-TK_YhDKQ_WHwnnTYks5sJQCzgaJpZM4OKoB2> .

[macroxue]

The optimal moves for some very difficult situations can be pre-calculated with exhaustive Expectimax and looked up instantaneously during the search. I wrote a controller based on this idea a couple of years ago. It achieved the 65536 tile 3% of the time. Since its search part was just simple Expectimax, your controller can probably do better with more sophisticated learning and the same lookups.