Overview

A solver for an extended version of Wordle with a word pool of 14,885 possible answers — far larger than the standard 2,315 — making naive frequency-based approaches less effective and requiring a more principled strategy.

Approach

The solver uses information theory to select guesses that maximise expected information gain — i.e. each guess is chosen to eliminate as many remaining candidate words as possible in expectation, regardless of whether it’s a likely answer itself.

matplotlib is used to visualise the distribution of guess counts across test runs, and collections for efficient candidate word tracking.

A torch-based component explores learned guess ordering as an alternative to the purely analytic approach.

Key Challenges

• The 14,885-word pool is large enough that brute-force entropy calculation over all candidates is expensive — optimising the search was necessary to keep guess selection fast
• Balancing exploitation (guessing likely answers) vs. exploration (guessing high-information words that probably aren’t the answer)

Tech Stack

Python · PyTorch · matplotlib · math · collections

---