yet I can't seem to come up with anything concrete for. Apologies if my problem is too trivial for this subreddit.
Due to my limited CS background (I've only had one class on general purpose programming and OOP so far, screw general education requirements), all I know about ML goes one way: given inputs, predict outcome (either regression or classification).
It took me a bit of thinking to figure out that what I'm trying to do is the exact opposite: given the desired outcome, suggest suitable inputs. A tangible example application would be playing a game (say, Poker) where the objective is to optimized winning chances, given as input the state of the game so far.
I can't do that. There are two problems:
If I train a set of parameters to map the input (presumably all the moves from previous games) to the output (win/loss/tie), the best I can do is predict my own doom in the middle of a match. Conversely, I can't train anything to map from the match's outcome back to how it should be played. I'm thinking I'd have to try something like the recommender system's collaborative filtering, but chances are I don't even know what I'm talking about here.
Even if I could reverse the engine, I will have to take as input the state of the game thus far to determine the next move. How do I get around an input of variable size?
[link] [6 comments]