SpellBond

Gym environment for the game Wordle. A simple action, state, and observation is included, with a random agent. Note that the spaces are very simplistic, and only included for a working example of the env.

How to run

• Python >=3.8 required
• Create python virtual environment and install the requirements.txt file:

python -m venv venv
source venv/bin/activate
pip install -r requirements.txt

• From the parent folder, install SpellBond as a package:

pip install -e .

To execute inference on full vocab, run the following command from scripts folder:

cd scripts
python run.py --task infer --test-file <path to testfile>

The test-file has a default value of "words.txt". Optionally, to execute inference on partial vocab size, run the following:
NOTE: Before executing, update the test-file to contain words only from this partial vocab list

python run.py --vocab-size {vocab_size} --task infer --test-file <path to testfile>

vocab_size can either be None (default, full vocab) or 1000.

Information regarding the algorithms implemented:

• Wordle_RL class represents the ACLA RL algorithm (adapted CACLA to discrete state/action spaces)
• SARSA class represents the actor-critic RL algorithm which can be trained either on-policy or off-policy