k78ma/connections-solver

Solving NYT Connections with vector embeddings.

NYT Connections Solver

This project aims to play the NYT Connections Game using vectorized embeddings of words.

The idea is to use embedding similarity as a metric for how good a given 4x4 grouping of 16 words is. This metric is then used as a fitness function for a genetic algorithm to find the best grouping, which hopefully allows us to avoid having to evaluate all ~2.6 million possibilties.

Requirements

• Gensim (pip install gensim)
  • You may need to downgrade scipy to a version below 1.13.0 for this to work.
• DEAP (pip install deap)
• numpy (pip install numpy)

Usage

• The script is written to run by taking a list of words in a .txt file in the puzzles/ folder.
• Each word in the file is separated by a newline. For an example, see 4-18-2024.txt.
• Then, the script can be run by python solver.py [filename]. For example:

python solver.py 4-18-2024.txt

TODO

• Add evaluation against actual solution