search

carykh / PrisonersDilemmaTournament

Watch This Place's awesome video about iterated Prisoner's Dilemma for context! https://www.youtube.com/watch?v=BOvAbjfJ0x0
MIT License
205 stars 159 forks source link

I quit. I have no chance. #50

Closed Barigamb738 closed 3 years ago

Barigamb738 commented 3 years ago

I just saw @l4vr0v's fork of the project with a bunch of other strategies and I was excited because I could test better. but it was sad to see my consistently winning strat fall down to 37'th place. So I don't really think I have much chance. I wish the best of luck to all of you guys.

ReimarPB commented 3 years ago

Just tried it myself, mine went down to 36th. It really shows how different the actual tournament is gonna be from the sample strats.

Barigamb738 commented 3 years ago

Yeah, I drastically overestimated my algorithm.

Quadrapod commented 3 years ago

It looks like my strategy is still able to take first here but I will say just because you don't think you'll win doesn't mean there's no value in competing. These are all pretty well refined scripts which are heavily optimized for that pool of competitors. The actual contest will be different and really it's not impossible for something like tit for tat to win in the end. It wouldn't even be the first time it's happened.

l4vr0v commented 3 years ago

Just for the record, some of the strategies- including the really good strategies with PascalCase names- in my repo are made by @Lasermancer. I included them in my repo (at his request) to test their performance in a competitive meta back when he was not yet familiar with git and GitHub. I will remove them now. I am not dominating this myself, and my own submission (npnntt) is likely not going to be very competitive. (@Lasermancer, @nobody5050, and I have been collaborating fairly closely and our strategies tend to riff off of each other's. I highly recommend doing something similar with [internet] friends. You will find yourself developing really cool strategies!)

That aside:

  1. Some strategies are very meta-sensitive. Compare the performances in my repo vs. the performances in the broader Prisoners-Dilemma-Enjoyers repo which has even more strategies.
  2. With a few key insights (forgiveness, opportunistic defection, minimal and effective punishment), your strategy can rocket up in performance. npnntt still does decently in some metas and that strategy is just "tit for tat, but cooperate if they defected after my defection unless they've never cooperated with me before."

Don't give up hope! You can get very very far.

Barigamb738 commented 3 years ago

Thanks guys. I will give it another shot.