search

pmariglia / showdown

A Pokemon Showdown Battle Bot written in Python
GNU General Public License v3.0
261 stars 178 forks source link

A few suggestions #55

Closed ghost closed 3 years ago

ghost commented 3 years ago

here are a few suggestions from my experience of this script:

  1. make nash-equilibrium run without docker. the docker took like 2 gb so making it run without docker would be appreciated.
  2. it doesn't switch pokemon. when its a in a sticky situation in a fight, it doesn't switch pokemon making the pokemon die.
  3. making it a neural network for extra intelligence???
  4. add a stall feature. I noticed when the bot was being slow due to other applications, the timer would be on and people would leave if no move would be done increasing my elo a lot. Overall the bot is amazing ,very fast and works very well. Thanks in advance
pmariglia commented 3 years ago

  1. Running without Docker requires the user to set up their own Gambit binaries - it's more of a headache to do that than to run Docker. Yeah the Docker image is based on an ubuntu image (reference) - the image size could probably be brought down by using something like Debian-slim, but not by too much. I can make that change when I get the chance but if you want to run without Docker you'll need to compile the Gambit binaries on your own machine, and then use this environment variable to point to the gambit-enummixed binary.

  2. At all? If it isn't considering switches then that is a bug, but you'll have to provide more context/evidence. Or does it just not switch in situations where you think it should switch? The bot isn't perfect - far from it and I never claimed it was. The decisions are just based off of the algorithm being used.

  3. I do not have the background knowledge nor the interest in doing this right now.

  4. Lol?

ghost commented 3 years ago

yea i am serious about the 4th one lmao

merkrafter commented 3 years ago
  1. On my (linux) machine, the Docker image "only" takes 1.3 GB. Are you on Windows? Anyway, I created a PR (#57 ) that saves ~600MB by utilizing multi-staged docker build. Maybe this is enough for your use case to profit from Docker. The rest of the size does not really stem from the OS image, but rather from the heavy Python libraries as pandas, scipy, etc. as far as I can tell.
pmariglia commented 3 years ago

Closed as none of the items are really addressable.

@merkrafter helped with this comment: ae34aa85ca0766a88b25e663d2e8735ad9094bfc which reduces the size of the docker image to just under 400MB (on my machine anyways).

The base image also uses python3.6-slim rather than Ubuntu now.