A Pokémon battle-bot based on NLP that can play battles on Pokemon Showdown.
Developed and tested using Python 3.6.3.
Environment variables are used for configuration which are by default read from a file named .env
The configurations available are:
BATTLE_BOT: (string, default "gptj") The BattleBot module to use. More on this below
TIMER: (string, default OFF) Request initiating timer at the beginning of the battle or not
SAVE_REPLAY: (bool, default False) Specifies whether or not to save replays of the battles
LOG_LEVEL: (string, default "INFO") The Python logging level
WEBSOCKET_URI: (string, default is the official PokemonShowdown websocket address: "sim.smogon.com:8000") The address to use to connect to the Pokemon Showdown websocket
PS_USERNAME: (string, required) Pokemon Showdown username
PS_PASSWORD: (string) Pokemon Showdown password
BOT_MODE: (string, required) The mode the the bot will operate in. Options are "CHALLENGE_USER", "SEARCH_LADDER", or "ACCEPT_CHALLENGE"
USER_TO_CHALLENGE: (string, required if BOT_MODE is "CHALLENGE_USER") The user to challenge
POKEMON_MODE: (string, required) The type of game this bot will play games in
TEAM_NAME: (string, required if POKEMON_MODE is one where a team is required) The name of the file that contains the team you want to use. More on this below in the Specifying Teams section.
RUN_COUNT: (integer, required) The amount of games this bot will play before quitting
ROOM_NAME: (string, optional) Optionally join a room by this name is BOT_MODE is "ACCEPT_CHALLENGE"
BANANA_API_KEY: (string, required if BATTLE_BOT is yamper) The API key to connect with the GPT-J model developped by Banana.
OPENAI_API_KEY: (string, required if BATTLE_BOT is gpt3) The API key to connect with the GPT-3 model developped by el tito Elon
There is a sample .env file in this repository.
1. Clone
Clone the repository with git clone https://github.com/gameduser/Yamper.git
2. Install Requirements
Install the requirements with pip install -r requirements.txt.
Be sure to use a virtual environment to isolate your packages.
3. Run
Run with python run.py and the bot will start with configurations
specified by environment variables read from the file named .env
This requires Docker 17.06 or higher.
1. Create the configuration file
Create an .env file like this example:
BATTLE_BOT=gptj
TIMER=OFF
SAVE_REPLAY=False
LOG_LEVEL=INFO
WEBSOCKET_URI=sim.smogon.com:8000
PS_USERNAME=<Your Pokemon Showdown user>
PS_PASSWORD=<Your Pokemon Showdown password>
BOT_MODE=SEARCH_LADDER
POKEMON_MODE=gen8randombattle
RUN_COUNT=1
BANANA_API_KEY=<Your Banana GPT-J API key>
2. Run the docker image
docker run --env-file .env gameduser/yamper
use BATTLE_BOT=safest
The bot searches through the game-tree for two turns and selects the move that minimizes the possible loss for a turn.
For decisions with random outcomes a weighted average is taken for all possible end states. For example: If using draco meteor versus some arbitrary other move results in a score of 1000 if it hits (90%) and a score of 900 if it misses (10%), the overall score for using draco meteor is (0.9 * 1000) + (0.1 * 900) = 990.
This is equivalent to the Expectiminimax strategy.
This decision type is deterministic - the bot will always make the same move given the same situation again.
use BATTLE_BOT=gptj (default unless otherwise specified)
Using the Banana GPT-J public model, it will decide each next movement by asking the model with an API call. Options are usually attacking or switching Pokemon, so first it asks what does it prefer. You will need to provide your own API key.
use BATTLE_BOT=gpt2
Using a GPT-2 public model, it will decide each next movement by asking the model with an API call. Same behavior as GPT-J.
use BATTLE_BOT=gpt3
Using a GPT-3 public model, it will decide each next movement by asking the model with an API call. Same behavior as GPT-J. You will need to provide your own API key.
The bots in the project all use a Pokemon battle engine to determine all possible transpositions that may occur from a pair of moves.
For more information, see ENGINE.md
You can specify teams by setting the TEAM_NAME environment variable.
Examples can be found in teams/teams/.
Passing in a directory will cause a random team to be selected from that directory.
The path specified should be relative to teams/teams/.
Specify a file:
TEAM_NAME=gen8/ou/clef_sand
Specify a directory:
TEAM_NAME=gen8/ou