Seeking: hints for starting with Genetic and/or Monte Carlo

AI Battles
#1

Hi! First of all, thanks CodingGame and everyone participating in it for such a great place for learning and having fun.

I participated in a couple of contests so far, did not have too much time to dedicate but had fun indeed. I always coded some deterministic heuristics with results that were… not really encouraging. It’s time to go to… the next level.

I think it might be beneficial for others in my condition to get hints from the gurus about:

  • what is the best (simplest?) Bot Programming competition to start with either Genetic Algorithms or Monte Carlo?
  • what do you recommend as a reading for getting started with the solution of the suggested competition in the bullet above? I know about the interesting tutorial on Genetic Algorithms by Sablier on Tech.io, it’s just that so far it was a bit difficult for me to bridge it to the challenges I competed in.

Thanks to anyone!

3 Likes
#2

that got to be Mars Lander!

1 Like
#3

Hi,

  • GA: it used to be CSB, but being old without a referee you’d have to reverse-engineer the game engine which is quite some pain even with tutorials such as Magus’s one: http://files.magusgeek.com/csb/csb_en.html
    So maybe MM would be better now.
  • MC / MCTS: I’d advise UTTT (easy engine, light game state, 81 turns max)
  • MinMax: WW
  • Getting started: Wikipedia + google “{algo} {language} tutorial” :slight_smile:
2 Likes
#4

Except that it’s not a bot programming multiplayer competition :slight_smile:
But yes it’s a great one to start with, it’s simpler since it removes the need for “dummies” to predict the opponent.

#5

Thank you very much! It took me a bit to figure out what UTTT but then… Google was my friend. As I understand it (and for others):

CSB: Coders Strike Back
MM: Mean Max
UTTT: Ultimate Tic Tac Toe
WW: Wondev Woman

2 Likes
#6

Thanks a lot! I was trying to solve that via heuristics but to little avail so far. I found a Javascript animated solution but I have probably misread it as “using GA is overkill for this puzzle” :smile:

#7

Indeed you got it right, sorry we’re so used to the acronyms :slight_smile:

#8

I’d still recommend starting with CSB over Mean Max - there are plenty of resources on how the physics work and it’s much simpler (you can get decent results ignoring shield / boost). Mean Max is quite complicated and due to the number of units / effects / edge collisions you have to have quite an optimized simulation to get good results.

One thing to note if you do start with CSB - get to gold league before trying to implement a simulation solution. The input format in earlier leagues makes it impossible to accurately simulate the game as you don’t know the velocities of units (and you can only calculate it if no collisions occurred).

1 Like
#9

The CSB Referee can be found here.