|
the roles defintiely influence the AI - no fear there... i am certain of that becuase of what i do in the playoffs with my bullpen... i was sick of seeing the bottom 2 guys used more than they needed to so i set them to unspecified roles. that seemed to do the trick. now, like clockwork, my top 3-5 guys get all the work in the playoffs.
there's likely a few more thngs going into that AI decision than just the depth chart and game situation. the ai likely has processes that help avoid tired bullpens with compact schedules. anytime it deviates, there is a cause for why it made that decisions (not saying it's a good decision or bad - not important to this context of the paragraph). they definitely need to continue to tweak the AI for the pen...
a basic decision tree is very simple thing for this situation but... i had to remove this paragraph because this makes sense:
i think the bigger problem is how to incorporate something like "high leverage" and "use more/less" and coaching preferences into the loop of logic. using a pecking order, checking fatigue and possibly the next X-days on schedul until a day off is not complex to handle and use for the pen despite what it sounds like... that other crap influencing it is the difficult part and i'd wager that "extra stuff" cause for bad ai decisions (not just preference, but stupid decisions, hands-down). that extra stuff gives more realism, though, so i am not saying it should be removed but maybe simplified in it's application... maybe 3-4 base decisions trees for overall bullpen use for certain types of underlying factors (the coach? more?), then adjust for other factors in the loop of logic that are more easily incorporated and applied well.
that, instead of all outside factors influence 1 decisions tree (or howewver you wish to visualize the logic). multiple bullpen decisions trees for various coach types as a base (or other factors if a better foundation) and then further manipulation within each of those exclusive decisions trees based on use more/less, higher leverage etc.. i think if they layered it differently it should handle all the qualifications better.
everything you do to make a decision, the ai has access to the same info - anytihgn relevant to the game.. some humans will use irrelevant things, lol. the ai can make decisions just like you do, if coded properly (obviously i mean 1 strategy and some variants of it, and not all of our indidivdual wishes as to how it should work). they just need to approach it differently if thy say it's "too hard"... it's just being attacked in the wrong way or wrong order... if it doens't work, stop doing it and do it differently. quit forcing the square peg through the round hole, lenny!
Last edited by NoOne; 12-30-2016 at 03:48 PM.
|