alt-gym-wordle 0.7.15

OpenAI gym environment for training agents on Wordle

gym-wordle

A wordle environment for openai/gym

Installation

Install openai-gym

then install this package with pip install -e .

Usage

import gym 
import gym_wordle
env = gym.make('Wordle-v0')

See the docs for more info

Environment details

This environment simulates a game of wordle using a wordle python clone from https://github.com/bellerb/wordle_solver

The action space is a discrete space of 12972 numbers which corresponds to a word from a list of all allowed wordle guesses and answers The observation space is a dict with the guesses and colors for the current game. Guesses is an array of shape(5,6) #5 letters and 6 rows where each element is a number from 0-26 where 0 is '' and 26 is z Colors is an array of the same shape, only each element is a number from 0-2 where 0 is a blank (or black or grey) square, 1 is a yellow square, and 2 is a green square

The reward calculation is as follows:

The agent gets 1-6 points depending on how fast the agent guesses the word. For example, getting the word on the first guesses rewards 6 points, getting the word on the second guess rewards 5 points, etc

The agent also is rewarded for colors in the current row (so the current guess). Right now, the agent is rewarded 3 points for each green tile, and 1 point for each yellow tile. No points are giving for grey tiles