I am learning about Bayes Networks at Stanford's online AI class. I was wondering if there are books that talk about and implement code related to Bayes networks in context of games?
Thanks
Bayes Networks in games?
There's not a lot of writing about Bayes in games because there's not a lot of use for Bayes in games. That's unfortunate because it is a really powerful technique. Most of the reason that we don't need to process partial information in the game realm is because, with regard to the game engine and entities, we already have ALL the information. What we can use it for is stuff that is not inherently a part of the game -- the player, for example. There is some use of Bayes to intuit player intent. For example, given a list of criteria about a situation, if the player is holding his sniper rifle, we are undetected by the enemy, we are in cover, and the player starts to move forward, chances are the player is not planning on rushing the enemy -- he wants to pop a melon or two.
That being said, in the book, AI Game Programming Wisdom, there is a great explanation with examples of Bayesian networks by Paul Tozour.
That being said, in the book, AI Game Programming Wisdom, there is a great explanation with examples of Bayesian networks by Paul Tozour.
There's not a lot of writing about Bayes in games because there's not a lot of use for Bayes in games.
I disagree. Dave, have you ever played Monopoly for the iPhone? <shudder>
It has 3 difficulty levels. At level 1 (easy) you can offer the computer the list price for a property it owns, and it will sell it. At levels 2 and 3 I'm pretty sure that the computer cheats-- as in the dice are loaded with magical weights that allow the computer to alter its 'luck' level. lol. It doesn't look like EA made any effort to build a 'smarter' opponent and IMO the game seriously suffers because of it.
Bayes would have been perfect for Monopoly.
So, OP, there is plenty of room for improvement in games, provided you're thinking about the right sorts of games.
Like I said, there are game for which it would be a powerful technique. Discerning the results of something stochastic (e.g. Monopoly) is one. I also seem to remember our discussion regarding the very Stratego-like game a few months back where I suggested that Bayes is the way of approaching that. However, in the face of actual perfect knowledge that the computer has, AI programmers don't need to use Bayes when we can just look up the info and fake stupidity. That's not my preferred method... I was just stating that it is often done that way.
A good example would be finding the player when he is hidden. Obviously the computer knows where the player is and can simply feign stupidity. However, that doesn't always lead to convincing error. Using Damián Isla's occupancy maps, however, is a Bayes-like treatment of influence maps. THAT leads to a very realistic approach by the NPCs because it is based on that magic combination of observation, belief, and possibility.
A good example would be finding the player when he is hidden. Obviously the computer knows where the player is and can simply feign stupidity. However, that doesn't always lead to convincing error. Using Damián Isla's occupancy maps, however, is a Bayes-like treatment of influence maps. THAT leads to a very realistic approach by the NPCs because it is based on that magic combination of observation, belief, and possibility.
Author
I am kinda disappointed finding out more and more(I am student by the way) that the jobs available in computer science are not that interesting as I thought. The things we study in theory, the cool stuff, don't have much applications yet. It seems the real world is way way behind in relation to the theory that is developed. I am not like oh I should quit computer science, far from that, I am just saying theory is way way too far from what is actually done in real world and what you get paid for.
I am kinda disappointed finding out more and more(I am student by the way) that the jobs available in computer science are not that interesting as I thought. The things we study in theory, the cool stuff, don't have much applications yet. It seems the real world is way way behind in relation to the theory that is developed. I am not like oh I should quit computer science, far from that, I am just saying theory is way way too far from what is actually done in real world and what you get paid for.
Calculemus, about 1% of people get the really cool jobs (like test driving Bugati Verons, reverse engineering fly brains, or designing video games). Everyone else is stuck doing things that your average teenager could care less about.
EA, in terms of a profit model, did the right thing by building a cheap AI. They wasted no money, and the 'Monopoly' brand name sold itself. The lame AI probably didn't affect sales one bit.
Dave is quite right in pointing out that most RTS or FPS games don't require an AI any more complex than a set of scripts. There really isn't much value in building complex AI for for most NPCs.
It has been my experience that if you want to do interesting cutting-edge things you have two options: Work for yourself, or work in academia. Build a game that uses a strong AI as the gimic that gets people to play it. It's never been a better time for the indy developer.
While it's true that intellectual products like new algorithms don't take huge numbers of people to produce -- so you'll be facing some competition to get these jobs -- I wouldn't totally write off these technologies as useless.
- Android and iOS devices both include speech recognition engines. Witness the much-hyped Siri. As far as I know, Hidden Markov Models are at the heart of most speech recognition algorithms; it's possible that Bayes nets are in use here too. And the speech recognition engines, though impressive, still leave much room for improvement. That means room for competition.
- The government funds a great many contractors that you've never heard of, who are trying to develop both old-fashioned database systems, and fancier statistical analysis tools including semantic networking tools, for understanding intelligence data. I imagine Bayes nets either get used here, or could be used.
- Robotics is very slowly taking off, not just in defense and in a few "silly" consumer applications like the Roomba (though it has some sophisticated competitors that even do SLAM!), but also in warehousing and factory automation, and we're just beginning to see the very leading edge of agricultural robotics. The world won't need a billion roboticists, but it is one more area where these sophistical tools can actually be useful.
I also think that games have a lot to teach these other fields. Sure, they don't need to deal with uncertainty to the same extent, but one thing they do a great job of is producing usable interfaces for interacting with the real world (or simulations thereof). People are beginning to acknowledge, for instance, that Starcraft is a pretty good model for what a good "net-centric warfare" interface should look like. Indeed, it was by explicitly following a strategy of copying Starcraft's UI that Ed Olson and his students won the recent MAGIC robotics competition in Australia. The difference, of course, is that instead of loading a map file you're doing SLAM, and the "fog of war" is real! My point in bringing this up is to say that, although some of these algorithms don't get used in games themselves, they get used in other fields that involve many of the same things as game development.
- Android and iOS devices both include speech recognition engines. Witness the much-hyped Siri. As far as I know, Hidden Markov Models are at the heart of most speech recognition algorithms; it's possible that Bayes nets are in use here too. And the speech recognition engines, though impressive, still leave much room for improvement. That means room for competition.
- The government funds a great many contractors that you've never heard of, who are trying to develop both old-fashioned database systems, and fancier statistical analysis tools including semantic networking tools, for understanding intelligence data. I imagine Bayes nets either get used here, or could be used.
- Robotics is very slowly taking off, not just in defense and in a few "silly" consumer applications like the Roomba (though it has some sophisticated competitors that even do SLAM!), but also in warehousing and factory automation, and we're just beginning to see the very leading edge of agricultural robotics. The world won't need a billion roboticists, but it is one more area where these sophistical tools can actually be useful.
I also think that games have a lot to teach these other fields. Sure, they don't need to deal with uncertainty to the same extent, but one thing they do a great job of is producing usable interfaces for interacting with the real world (or simulations thereof). People are beginning to acknowledge, for instance, that Starcraft is a pretty good model for what a good "net-centric warfare" interface should look like. Indeed, it was by explicitly following a strategy of copying Starcraft's UI that Ed Olson and his students won the recent MAGIC robotics competition in Australia. The difference, of course, is that instead of loading a map file you're doing SLAM, and the "fog of war" is real! My point in bringing this up is to say that, although some of these algorithms don't get used in games themselves, they get used in other fields that involve many of the same things as game development.
Author
Thank you guys, I got a good idea about Bayes Nets in games.
In week 3 Machine Learning course, we learn about Neural Networks. So that I don't start a new topic, you guys wanna talk about Neural Nets in games? Are they used and to what degree?
Thanks
In week 3 Machine Learning course, we learn about Neural Networks. So that I don't start a new topic, you guys wanna talk about Neural Nets in games? Are they used and to what degree?
Thanks
Thank you guys, I got a good idea about Bayes Nets in games.
In week 3 Machine Learning course, we learn about Neural Networks. So that I don't start a new topic, you guys wanna talk about Neural Nets in games? Are they used and to what degree?
Thanks
Even the most basic search of this forum's posts will give you the answer(s) about NNs in games. Trust me... we don't want to restart that.
I am learning about Bayes Networks at Stanford's online AI class. I was wondering if there are books that talk about and implement code related to Bayes networks in context of games?
Thanks
Bayes Networks would be wasted on games. They are slow, inference on them is NP - hard. Building the network structure from data is also NP-hard. So for a decent sized network you gonna be doing something like gibbs sampling to do inference on network a structure that is almost guaranteed to be wrong. For a game the AI will take a long time to get at the distribution and is not worth it. Most of machine learning is either too slow, data intensive or stationary for Game use.
If there would be one machine learning thing that I would actively look into, it would be Decision Trees. Decision trees are brittle so can add variance and do well with little data. You could then augment them as Random forest or boosted trees. I think that would be where I would start.
A close cousin to bayes nets that might be useful to some types of games (arcade shooter, anything requiring movement tracking and prediction) would be a particle filter. Finally, a bared down Reinforcement learning algo may be good for long term play in a strategy or RPG. All of these would be very hard to get right and take a lot of time when a simple markov chain based model or even Finite state machine would have done just as good or better for much less work.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
