Hi,
well thats a very broad question
You say you don't have much experience in Jass, do you know other programming languages? Because many techniques are quite complicated...
If you want a simple decision model like the one posted in the thread, it would be the best to have a look at Finite State Machines. Its the most common (and easiest) approach to design an AI for a computer game and also widely used by professionals. The Hidden Markov Models can be seen as an extension to this approach. Here also the Principle Component Analisis becomes interesting.
For turn-based games there exists the Alpha-Beta Pruning algorithm which is used to design extremly strong AIs. This algorithm can't be applied to a real-time game like wc3, however it might be interesting to understand more about AI in general.
If you really want to enter machine learning, you should read about statistical learning techniques like Support Vector Machines, probabilistic ones like genetic algorithms and the Multi Layer Perceptrons. Those techniques are widely used for both regression and prediction as well as for decision making.
Considering a dynamic process like a running wc3 game, static feedforeward networks won't be enough, so you might take a look then at Recurrent Neural Networks. Those are the most powerfull machine learning tools available at the moment, however the theory behind them is quite complex.
I want to try to implement such things some day to wc3, however before one can start you would need a decent Math library which provides methods for nonlinear optimization and so on. I started with this library (Advanced Maths), so if you want to get started, you can take a look at it.
If this is all to much for you, you may just take a look at Finite State Machines. They are farely easy and allow you (with some tuning) to build acceptable AIs.
lfh
well thats a very broad question
You say you don't have much experience in Jass, do you know other programming languages? Because many techniques are quite complicated...
If you want a simple decision model like the one posted in the thread, it would be the best to have a look at Finite State Machines. Its the most common (and easiest) approach to design an AI for a computer game and also widely used by professionals. The Hidden Markov Models can be seen as an extension to this approach. Here also the Principle Component Analisis becomes interesting.
For turn-based games there exists the Alpha-Beta Pruning algorithm which is used to design extremly strong AIs. This algorithm can't be applied to a real-time game like wc3, however it might be interesting to understand more about AI in general.
If you really want to enter machine learning, you should read about statistical learning techniques like Support Vector Machines, probabilistic ones like genetic algorithms and the Multi Layer Perceptrons. Those techniques are widely used for both regression and prediction as well as for decision making.
Considering a dynamic process like a running wc3 game, static feedforeward networks won't be enough, so you might take a look then at Recurrent Neural Networks. Those are the most powerfull machine learning tools available at the moment, however the theory behind them is quite complex.
I want to try to implement such things some day to wc3, however before one can start you would need a decent Math library which provides methods for nonlinear optimization and so on. I started with this library (Advanced Maths), so if you want to get started, you can take a look at it.
If this is all to much for you, you may just take a look at Finite State Machines. They are farely easy and allow you (with some tuning) to build acceptable AIs.
lfh
