Re: An AI question and a UI question

> You threat both missing the target and beaing hit as penalty. In addition,
> all the monsters that can see each other (or communicate) should share
> the same neural network -- so that they can learn also from other's actions.

I think they should share the same learning set (collection of hits &
misses of the hero ). Then each monster is allowed to train  n times
on that training set. n depending on how fast a learner the monster
is.

> 
> Offcourse, once the player changes his weapon or advances a level, the
> optimal distance changes.  But it will take a while for mosnters to notice
> that.
> 
> Perfectly, the monsters would be able to evaluate the optimal distance
> also from success or failure of other monsters, not neccessarily having
> the same neural network. Not sure how to do it, though.

Yes, that's difficult. You'd have to used a 'perceived strength',
'perceived skill' etc. It's logical that Dob the Goblin misses the
hero, it's scary if StarSpawn keeps fumbling against the hero, any
monster less than a StarSpawn should run away.

> 
> You'd need only few neurons. If you want to include info on character's
> equipment.

The problem is that if you want the movement of monsters to be
governed by NN's, you will have quickly a large amount of neurons.
Note that this NN could solve the path way finding for you if you
train long enough.

T.