A man with a .plan!

diracdelta on 2011/09/11 08:16 of Dark Solitons
— edited on 2011/09/11 19:12

So, genetic algorithms. Bits describe connection weights for AI MM. 100 States seems adequate.

The player moves around a maze hunting blobs. The blobs will do something. If they collide with the player, they will harm him. Blobs come in waves? Or do they spawn directly off of 'successful' blobs?

(log in to comment)

Comments

nqe on 2011/09/11 14:07:

Oh man, if you manage to get this to work, it could be pretty awesome. I can already imagine myself trying to game the system by hunting only successful blobs (if there was a way to differentiate them) and leaving the weaker ones to spawn the next generation in turn improving my chances of survival. Even if I'm imagining a dynamic which won't exist, genetic algorithms are always fun to see how they evolve.

diracdelta on 2011/09/11 20:08:

Working on dfa.py

We're going to have a Markov Model be a descendent of a regular DFA with an overridden transition function (that essentially acts as a gate keeper)

Each Node in the DFA will keep track of:
  -Transition Functions
  -User Data

diracdelta on 2011/09/11 20:09:

Hey nqe,

Glad to hear it sounds interesting. :) The main problem I'm worried about, however, is that the player will get bored before the enough generations elapse for the blobs to exhibit anything interesting. :(

ehsanul on 2011/09/12 05:39:

diracdelta, you can sorta fix that problem by starting the player off with some interesting blobs you evolved yourself over many generations + random mutations.

However, part of the fun of genetic algorithms is seeing how they converge from completely stupid solutions, to slightly less stupid ones - so that would be missing if you do as I suggest.