Regular Expression Crossword Puzzle | Hacker News Reader

Regular Expression Crossword Puzzle

gregable.com

148 pointsgregable35 comments

Here's a nice site for these: https://regexcrossword.com/

The default puzzles are a bit basic, but there's a large amount of player-created puzzles that are great.

I love these! I tried automating the creation of them a while ago[1] (refreshing the page creates a new one). I should probably revisit it though as the generated puzzles aren't tricky enough..

[1] http://padolsey.github.io/redoku/

There's also an extension for this: https://github.com/wolfascu/regex-crossword-goodies

The original source for this puzzle was the 2013 MIT Mystery Hunt:

http://www.mit.edu/~puzzle/2013/coinheist.com/rubik/a_regula...

and it is credited to "Dan Gulotta, based on an idea by Palmer Mebane". (It's too bad when people republish puzzles without explaining where they're from or linking to the original.)

Awesome, thanks! I've added attributions on both the blog post and the puzzle page. It's a really fun puzzle, the author of it definitely deserves credit.

Thanks for updating your post with the attributions!

One note about the interface - the rotation arrows are very confusing. You should always put the curve on top, rather than on bottom, since most people think of rotation directions based on the movement of a point on the top of the arc. Additionally, the left arrow rotates right, while the right arrow rotates left.

I reversed the order of the arrows, but I'm not sure about flipping the glyphs. Those are the standard unicode glyphs for rotation arrows, I didn't want to use any images on the page. I suppose I could use CSS to invert them.

When the top moves left the bottom moves right.

I had a lot of fun writing a program to solve this puzzle last year. I wrote about it here: http://almostobsolete.net/regex-crossword/part1.html

The Haskell code is also on GitHub

Sites like http://regexper.com/#.*OXR.* help interpret the clues without solving the puzzles

I solved this a while back. It's a bit challenging, but mostly due to the fact that the clues aren't really completely determining the solution, so you can't deduce each square like you can in e.g. sudoku. You have to make some guesses and then backtrack if they're wrong.

FWIW, I found that you can either deterministically deduce a cell or prove that the choice is irrelevant for at least two of the conditions. In this case, you can fill it with the solution that gives you the most flexibility for the third.

It took me a while, but I solved it without any guessing+backtracking.

Same. Definitely doable. It did take me a guesstimated 1.5h though.

Agreed, you can definitely arrive at a deterministic answer. (takes a lot longer though.)

Assuming a puzzle had a single solution: How computationally hard would it be to prove that one solution was the only solution?

Could you move the puzzle on amazon a few (20?) pixels to the right, the.(.)(.)(.)(.)\4\3\2\1. regex gets cut off on the left when you rotate http://i.imgur.com/qcG1kav.png

I added 20 pixels of left margin to the puzzle, let me know if that helps. I think it depends on screen resolution, I'm not seeing the same thing. If it's still causing you trouble Cntrl +/- should work for resizing the page too.

It's fixed now. Thanks

After spending most of the morning solving it manually I was thinking about how an algorithm might work for this.

Perhaps converting each regex into a DFA, but there are too many dependencies with all the other regexes. Any ideas on how to avoid brute forcing it?

This kind of puzzle falls under a general class of problems called Constraint Satisfaction Problems. Sudoku puzzles, map coloring, and cryptarithms are all examples of CSPs. A CSP is defined by a bunch of variables, each of which can take on some values in their domain (In this case each cell is a variable, and initially their domain is all 26 letters.) and a group of arbitrary constraints over the variables (the regexes).

One method of solving a CSP is to keep a set that initially contains all of the variables. On each iteration you choose a variable from the set and remove all values in its domain that are impossible given the constraints containing the variable and the domains of the other variables in those constraints. Then you add all of those other variables to the set because they are now restricted more than before. You keep doing the above until you get stuck (which may or may not happen). You then start guessing. There are a bunch of heuristics about how you should choose which variable to guess (eg the variable with the smallest domain). Once you have made your guess, you then add all of the variables back to the set. If you find out the puzzle is impossible, you backtrack to the last guess you made. I believe there are ways you can determine which guess was the problem and immediately backtrack to there.

Seems like this puzzle might be a candidate for solving using Donald Knuth's Dancing Links algorithm. Setting it up, however, might be as much work as solving it manually.

What books come highly recommended for learning CSP?

I wrote my master's thesis in Constraint Programming. The textbook for my CP course was Apt's "Principles of Constraint Programming"[1], but I have to admit to barely using it. Most of the books and "introductions" to CP which I've read have been lacking in either rigor or clarity; many important details are glossed over, partly I think, because of CP's development from other kinds of optimization/satisfaction programming. When learning, I mostly used lecture notes from Christian Schulte (from KTH in Sweden) and Pierre Flener (from Uppsala University, also in Sweden). I can unfortunately not find these accessible online at the moment.

[1] http://www.amazon.com/Principles-Constraint-Programming-Krzy...

I've actually learned everything I know about CSP from chapter six of Artificial Intelligence a Modern Approach.

You can turn it into an SMT problem and solve it with a standard solver. In Haskell it's quite beautiful as you can write code that's basically identical to writing a parser, and then generate inputs that satisfy the parser magically.

https://github.com/ekmett/ersatz/tree/master/examples/regexp...

These regexes aren't really regular expressions. They're PERL regexes with backreferences so a DFA won't do.

My thought was to try ragel with the intersection operator, but for most of them it just result in something that'll match any character... maybe I'm doing something wrong.

How about that for a CAPTCHA. :)

CAPTCHAs should be easy to solve for humans and hard for computers, not the other way around :P

I did this puzzle with a coworker and regular "commits" ( photocopy and add a version number ) was pretty useful. We botched it more than once and it was nice being able to revert.

Great puzzle, too bad it doesn't have a single solution...

I'm fairly sure there is only one solution.

Reply on news.ycombinator.com