Quote:
Original post by willh
Quote:
Original post by InnocuousFox
Quote:
Original post by willh
How about time-series analysis?
So a guy who barely gets Minimax and A* should do time-series analysis as a next step? Uh... no. There are far more productive and learning-conducive things he could be doing.
It's a sad day for GameDev when the AI forum moderator thinks trying to model future events based on past events is not productive.
I believe that if you use your advanced AI skills to analyze my post (conveniently quoted above), you will note that I didn't say that time-series analysis is not valuable. (In fact, I have done a few n-gram tricks myself in the past -- although not necc. in game AI). What I DID point out, however, was that the OP is obviously new to the field. He doesn't seem to understand simple state transition decisions based on simple rule-based criteria. I think that he is not going to be well served by attempting a more complicated decision algorithm until he knows how to make a decision in the first place.
Your approach was similar to taking a 7 year old who is beginning to grasp multiplication (but can't quite make the leap to division) and suggesting that calculus would be the next logical lesson for him to learn. That is not to say that calculus, as a mathematical technique, is not productive... but rather that trying to
teach the hapless child calculus at this stage would not be productive for either the student or the teacher.
Quote:
Who would really be surprised though? Dave Mark is the same AI 'expert' who pre-flamed any and all people who dare ask a question about ANNs afterall.
Mis-represented and irrelevant to this discussion.
Quote:
The most basic form of machine learning is regression analysis...
Irrelevant to the OP and to the discussion. You sound like a salesman, by the way.
Quote:
If you're interested in predicting future events based on past behaviour, time series analysis is your friend.
And this brings us to the point. The OP isn't "interested in predicting future events based on past behavior." He's interested in writing a simple decision-making agent that needs to look at its own bloody context in the world. He doesn't even know how to make a context for the agent to look at. We're trying to help him with that. If he can't make a context for decisions, he can't reason on them. More importantly, he would have no framework for even recognizing, much less recording and analyzing the agents' actions.
Put another way, reacting to a player's behavior is all about turning knobs on decisions. He doesn't even have any knobs to turn yet.
Quote:
Why anyone would want to model behaviour in a video game is beyond Dave Mark though...
Game AI is 99% about modeling the agent's behavior -- not the player's behavior. In fact, you can't react intelligently to the player's behavior unless you can model the agent's behavior. Again... there's a lot of math to be learned before calculus.
Perhaps your own personal AI needs to work on an English parser that better understands the contextual relevance of other people's posts rather than the artificial extension of your own epeen?
