T1j Twixt engine by Johannes Schwagereit TWIXT PP

11 replies. Last post: 2020-12-07

Reply to this topic Return to forum

T1j Twixt engine by Johannes Schwagereit
  • David J Bush ★ at 2020-08-25

    For those who want to play a computer opponent not as strong as TwixtBot, you could always download T1j. It doesn't do link removal correctly, but it does offer the pie rule which is much more important, and better still, it offers row handicapping. So if it is stronger than you you could take a handicap, or if you are stronger you could give one.

    This has been available for a long time. It might require an older version of Oracle Java such as 10.0.2

  • David J Bush ★ at 2020-08-25

    Here is a link to Java 10.0.2 Don't install it, just extract it to a subfolder and use its full path when invoking it.

  • MisterCat ★ at 2020-08-25

    This IS nice of you to post for people, David.  I've had T1j for years, but rarely ever play it.  Compared to the old, WEAK, Twixtbot (which is worth around 1600 strength), I would give T1j only a rating of around 1200; again, that is in MY estimate.  It IS still worth having, and I'll admit that I have NEVER fooled around with the row handicapping feature - and perhaps I COULD arrange more of a challenge this way.

    There is an Android phone app (free download at Google Play) which plays Twixt, and the creator states that he is using the same algorithm as T1j, meaning it is not very good.  However, it DOES play the game, and can even WIN - if you are not careful.  Again, FOR FREE, what the heck - may as well download and install.

    I still want the ol' weak Twixtbot back!

    mc

  • TwixtBot at 2020-08-27

    If someone wants the source code to the old twixt bot and wants to polish it up, I can send it to you.  The new one too.  You can drop it down to ~250 trials and it makes a tough interactive opponent.

  • add3993 at 2020-08-27

    I would love to learn Twixt strategy from playing strong Twixtbot (on a PC, because the LG bot, while appreciated, is slow).  I just don't have time to struggle with source code, although I thank you for making it available on your github.  But wanted to note two design goals for anyone interested in improving user experience:

    1) aside from the obvious goal of a standalone .exe with gui, can it be integrated with jtwixt or any other Twixt game viewer?   https://canyon23.net/jgame/README_twixt.html

     2) Can the AI provide win% feedback to the player (or any other useful data from the neural net) to facilitate human learning?

  • TwixtBot at 2020-11-21

    1) almost certainly

    2) yes, it has a nice win % number.

  • PFW at 2020-11-23

    Ok

  • Hairspring at 2020-12-01

    To add3993's statement that he or she would love to learn Twixt strategy from playing strong Twixtbot I would say that trying to learn Twixt strategy from Twixtbot is futile. It just doesn't play the way a human does and most of what it does will appear to be a mystery if you do not already have a very good handle on how the strategy of the game works. I won't pretend to understand how the bot works (I am not even close to a code or math guy) but my limited understanding is that it can game out iterations of best moves in a chain pretty far out and ultimately when it plunks down a peg it has little resemblance to where most of us mere mortal humans might do so. It doesn't teach you anything about how to play against a human. You might learn how to program a machine to play against another machine which seems like fun if you like that kind of thing but unless you can spend an enormous amount of human time to match the iterations of code that the bot uses for each move you are unlikely to play like Twixtbot. It is a really cool and impressive set of ones and zeros though. Play the game, look at David's game analysis that he does sometimes, do some of the puzzles. Those are better learning tools.

  • bob440 at 2020-12-02

    T1j works just fine with the default java package on my 20.04 Ubuntu desktop.

  • bob440 at 2020-12-02

    My java version …

    > java -version

    openjdk version “11.0.9.1” 2020-11-04

  • add3993 at 2020-12-07

    Hi Hairspring—if I (he/his) can't learn basic strategy of such an elegant game, by playing what seems the strongest player, I'll consider it my own failure and not bear any resentment.  Learning by beat-downs is kind of a fun hobby in its own right, regardless of efficiency.  Also, it doesn't seem like there are enough lower-level puzzles available.  I just hope the bot will be made fully accessible to non-coders (and people like me who code in theory but are rusty and short on time).

Return to forum

Reply to this topic