When building a crossword layout, the generator makes a sequence of decisions. Each decision is greedily made to maximize a hard coded function.
The hard coded function incentivizes number of connections between words, longer words first, equal number of vertical and horizontal words, and word placement near the center.
Let's call the hard coded function our fitness function (even though it might not be a conventional fitness function).
Goal: Evaluate the performance of our crossword layout generator's fitness function.
Here are some things that we can do to help us meet our goal.
We can create a data set of professionally made crossword puzzle layouts.
We can propose approaches for vectorizing crossword layouts.
We can run simulations to try to learn fitness function parameters that lead to better crossword layouts.
When building a crossword layout, the generator makes a sequence of decisions. Each decision is greedily made to maximize a hard coded function.
The hard coded function incentivizes number of connections between words, longer words first, equal number of vertical and horizontal words, and word placement near the center.
Let's call the hard coded function our fitness function (even though it might not be a conventional fitness function).
Goal: Evaluate the performance of our crossword layout generator's fitness function.
Here are some things that we can do to help us meet our goal.