Code of Kutulu - Feedback & Strategies

BlitzProg 120th.

Took monday and tuesday off work for Codingame, which I ended up using to fix the computer of a friend and dealing with some social events. When I went back to work, everyone had a cold, and they gave it to me for rest of the challenge. I ran out of pokédollars so I couldn’t fix myself with a Full Heal.

Wood3 to Mid Silver : Quick AI to explore the early leagues.

I copy the map, put a fake wall on and around everything that’s not an explorer. I then have two cases:

  • If I’m on a “fake wall” and can move to a place next to me where there isn’t : move there.
  • Else : go to the last explorer of the list given.

Mid-silver to high gold : Monte Carlo with heuristic, Depth 6.

  • Didn’t have the motivation to finish my simulator completely, so it had prediction errors. In addition, I also unable to get any AI working for player dummies, because they would always be doing a misleading choice. So I ended up treating these as still ghosts that ennemies t̶a̶r̶g̶e̶t do not target. (edited)

  • Because of the above I had to treat predictions errors in a fail-safe way, meaning I would dodge a threat, even if I didn’t happen - However many players with better simulations would detect there was no threat and I would end up making choices that would separate me from the rest of the team.

  • Lots of adjustments were made to ensure I would prioritize being near an explorer, even if it meant I had to run in the path of a Wanderer.

  • Abilities are attempted whenever Monte Carlo decides to not move. The higher the sanity, the harder I try for better opportunity (like causing more damage with a Yell or using a plan on every players)

The contest was very interesting. I liked the idea of staying close to other explorers, it gave a co-operative touch to the contest. I wish to see more of that in the future :slight_smile:

5 Likes