Author |
Topic: Shannon Type B (Read 2948 times) |
|
tize
Forum Guru
Arimaa player #3121
Gender:
Posts: 118
|
|
Re: Shannon Type B
« Reply #15 on: Mar 22nd, 2009, 12:22am » |
Quote Modify
|
Quote:Tize, when you say that you use marwin's evaluation function for biasing playouts, do you mean you use the information from the function at the node to bias all the playouts, or that you use it at each ply (maybe evaluating all possibilities) ? In the latter case, playouts must be horribly slow ! And you would anyhow see less nodes than with alpha-beta... |
| During the playout carlos generates a list of all steps avaliable, evaluate these positions and add a noise. Then he picks the step with highest score.
|
|
IP Logged |
|
|
|
Hannoskaj
Forum Guru
Arimaa player #3794
Gender:
Posts: 75
|
|
Re: Shannon Type B
« Reply #16 on: Mar 22nd, 2009, 3:36am » |
Quote Modify
|
Quote: During the playout carlos generates a list of all steps avaliable, evaluate these positions and add a noise. Then he picks the step with highest score. |
| I don't know how fast is your evaluation function, but have you considered making one (much) faster and (much) less precise for biasing your playouts ? Maybe also for final evaluation, though less so. And I guess that the noise level is a very important parameter...
|
|
IP Logged |
|
|
|
|