This repository contains code that creates questions and answers for a topic that you supply. It uses the question-answer pairs to create a crossword puzzle.
You supply an initial prompt. The code searches a vectorized collection of Wikipedia data and generates a list of people and places that are associated with the topic. The list is a list of potential puzzle answers.
Once it has the list of answers, the code searches the Wikipedia collection again to create a clue for each answer.
Finally, the code passes the list of clues and answers to a puzzle generator. The puzzle generator creates a crossword puzzle that uses as many of the answers as it can fit into a puzzle grid.
There are two notebooks in the juypter-notebooks directory.
Use crossword-setup.ipynb to create a collection of data based on Wikipedia.
The notebook uses the Simple English subset of the Wikipedia data set from Cohere.
This is where the language is configured: import_wiki_data("simple", num_rows, skip_rows).
To import a different language, uncomment one of the other import rows or edit the code to add the language name.
Use crossword-create-puzzle.ipynb to generate the puzzle.
The example code creates a puzzle based on the Harry Potter series. This is the
query prompt: query="Get the name of a Harry Potter character or the name of a place associated with Harry Potter" To change the puzzle topic, edit the query
prompt.
This project uses Cohere for the data set and LLM. You
need a Cohere API key to run the code in crossword-create-puzzle.ipynb.
The code that generates the list of clues and answers is new.
The code that uses the list to generate a puzzle is a lightly modified version
of the crossword_helmig
project.