I still think df(t)/dt is a function of ROI(tau) where tau ranges from 0~t, (see my previous question), in other words, LOCAL optimization is NOT GLOBAL optimization. Therefore your formula about maximizing df/dt is not valid at all.
It is more appropriate to use discrete Markov chain (MC) to model the civ game, although the state space may be huge.
In the language of MC, you need to find a globally optimized path from turn 1 to turn ~300, but your approach only concerns about the "best" next state.
Well, I do not oppose that your model may yield a good "path", but in my point of view, the model based on ROI is just too simple and sometimes may misleading.
[ 本帖最后由 binary 于 2010-2-18 09:56 编辑 ] |