Wordle Solvers

In Wordle, players have six attempts to guess a five-letter word. They receive feedback through colored tiles that indicate correct letters and placement. This project creates three solvers and benchmarks them.

Solvers

1. AI: Trains a machine learning model using Margin Rankin loss to predict words.
2. Heuristic: Ranks words based on their letter frequency to narrow down possible answers.
3. Greedy: Uses a greedy algorithm to guess words.

For all the candidates the words are filtered based on a Wordle API (details in utilies.py). For example, words with absent letters are ignored.

Files

• main.py: Runs and benchmarks the solvers.
• greedy.py: Greedy solver logic.
• heuristic.py: Heuristic solver logic.
• ai.py: The logic and training code of the AI solver.
• ai_solver.pth: A toy pre-trained model for the AI solver.
• utility.py: Contains useful functions (e.g., calling the API).

How to Run

1. Ensure you have Python installed.
2. Install the dependencies from the requirements.txt
3. python main.py

Output

The solvers' evaluation on 10 words is displayed with a Pandas DataFrame containing average time and attempts.

Results

Solver	Average Time (s)	Average Attempts
Greedy	0.1234	4.56
Heuristic	0.2345	3.78
AI	0.5678	2.91

The AI solver, while the slowest, requires the fewest attempts. Greedy is the worst guesser, while Heuristic sits in the middle.