Pluribus poker preflop defense frequencies

The way Pluribus defends pre-flop is interesting; we’ll do a few posts on it, for now let’s focus on poker preflop defense frequencies.

pluribus poker preflop defense
average size in X times previous bet (x time rfi)

Key learnings on preflop defense strategy

  • Overall in position defense frequencies are very similar to the pros frequencies and “current wisdom”
  • Maybe a bit less 3B and more call from the button
  • Significant differences from the blinds
    • Pluribus defends a lot more from the small blind, with a lot more calls
      • like twice as much as the pros did in this game!
    • Pluribus defends less from the big blind, will less calls
    • 3Bet sizes from the blinds are also larger than the pros and the averages we see online.

Preview on preflop defense hand ranges in position

I’ll need a full detailed post on this but many “new” behaviours to me in position:

  • Very mixed strategy between call and 3bet with the same hands
  • Some fairly surprising hands to 3bet:
    • pairs all the way down to 55
    • KQo and ATo at high frequency
    • A lot of suited connectors, but that’s similar to Snowie

Further study needed

I need to look at more detail, ie position of the original open, size of the open and whether these have impact or not.

I only looked at average 3bet size, not the distribution of these, don’t know yet whether there is a big spread.

I also haven’t looked at ranges from the blinds.

So there is a lot more coming.

2 thoughts on “Pluribus poker preflop defense frequencies”

  1. Here are notes from one of the players:

    This is Jason Les, a pro who participated in the challenge.

    I do genuinely believe that Libratus and now Pluribus have been the first AI breakthrough on their respective games. I believe there have been winning bots for a very long time, but they have had to be selective on their games. I.E.: play low stakes. I am certain I encountered many bots before playing Libratus, and I crushed them. It was some of the best money I ever made. I just logged on every day and this dumb bot played me for 12 hours. So Libratus and Pluribus are breakthroughs because they are better than all human opponents, not some (or “most” if you want to be generous).

    I don’t have a ton of info on OBORRA because he/it became a thing around the time I stopped playing online. However, a player like “40and7” I think its a similar case. Most likely was a human getting some real time assistance, and seemed to me to have some significant limitations on stack size. If I recall correctly, he couldn’t really play deeper than like 130bb? I could be wrong.

    Hi I’m Jason Les, a poker pro who participated in the challenge.

    Preflop: Played looser preflop than most humans, but 3bet less and folded to 3bet more.

    Postflop: It was more aggressive in multiway pots than humans typically are. Ex: 4-way to the flop and it just jams a low FD+GS over a cbet.

    https://www.reddit.com/r/MachineLearning/comments/ceece3/ama_we_are_noam_brown_and_tuomas_sandholm/

  2. Notes by one of the players:

    This is Jason Les, a pro who participated in the challenge.

    I do genuinely believe that Libratus and now Pluribus have been the first AI breakthrough on their respective games. I believe there have been winning bots for a very long time, but they have had to be selective on their games. I.E.: play low stakes. I am certain I encountered many bots before playing Libratus, and I crushed them. It was some of the best money I ever made. I just logged on every day and this dumb bot played me for 12 hours. So Libratus and Pluribus are breakthroughs because they are better than all human opponents, not some (or “most” if you want to be generous).

    I don’t have a ton of info on OBORRA because he/it became a thing around the time I stopped playing online. However, a player like “40and7” I think its a similar case. Most likely was a human getting some real time assistance, and seemed to me to have some significant limitations on stack size. If I recall correctly, he couldn’t really play deeper than like 130bb? I could be wrong.

    Preflop: Played looser preflop than most humans, but 3bet less and folded to 3bet more.

    Postflop: It was more aggressive in multiway pots than humans typically are. Ex: 4-way to the flop and it just jams a low FD+GS over a cbet.
    ______

    Source:

    https://www.reddit.com/r/MachineLearning/comments/ceece3/ama_we_are_noam_brown_and_tuomas_sandholm/

Leave a Reply

Your email address will not be published. Required fields are marked *