Look this: ________________________________________ (#define-learning-object MYOBJECT (activate-for same-map-size) (activate-for same-player-number) (activate-for same-map-type) (define-number-options 3) ) ; then after in the AI script #load-by-learning-object MYOBJECT option 1 (your code here) #load-by-learning-object MYOBJECT option 2 (your code here) #load-by-learning-object MYOBJECT option 3 (your code here) ________________________________________ So this would work in this mode: when your AI plays a game with certain settings, the result of the game would be saved in a log file. So if your AI has played 10 games with map-size=tiny, player-number=2, map-type=coastal, the result of that 10 games would be remembered with something like: option 1 => 3 played 2 won 1 lost option 2 => 3 played 1 won 2 lost option 3 => 4 played 3 won 1 lost So the preprocessor command "#load-by-learning-object MYOBJECT" would load something like ( for example): 30% chanches option 1 20% option 2 50% option 3. We now suppose that the AI will load option 1 and win the game. So next time, option 1 will have 45% chanches to be loaded and other option less than their actual 20% and 50% And this is the only way i see to have something similar to a "learning AI". For example if in the first option AI plans a knight rush, in the second an archer+phanalx mass attack and in the third to build defenses and a big economic boom, the learning objetc before or after will be able to choose more likely the strategy that you human player aren't able to counter.
