"When the autosolver is confused, it will guess and pick a random tile, usually failing" Realistic.

For more effective screensaving, you'll want to randomly offset the lines so they aren't in the same place every time. Possibly also lower the contrast. For LCD screens, you'll probably want a light theme, as white pixels are the "relaxed" state for LCDs (AFAIK). For OLED, you'll want a dark theme.

I knew someone who coded themselves out of games. This was around when Sudoku first became popular--he basically said "if I can fully solve Sudoku by programming a solver for it, I've effectively solved every Sudoku ever, and I never have to play the game again. Same for every other solvable game." (I assume he was at the time asked a lot if he played lol)

To improve the style of the code, I have a few simple suggestions, you can improve the indentation wall by refactoring the if condition statement to if not condition return/continue. Also, the part that checks for nearby cells can be refactored to iterate over a delta vector and extract it into a function, which seems to be used in many places. Happy coding :)

Actual great hook, made me want to watch the entire video from the start. Please make more videos!

When i get to ambiguous areas, i find probability analysis useful to increase my chances

4:31 The solver appears to have revealed a 6, revealed a tile that was obviously a bomb as there were only 6 unrevealed tiles next to the 6, and then proceeded to flag another tile before the solver's failure fully registered.

I sometimes played this game, but ultimately when I actually had the patience to try solve a hard board, I always ended up in a situation where I had to guess, and always got it wrong...

Amazingly high quality for a channels second vid. I would gladly watch more videos of you coding things.

3:20 that 3 in the 3rd row on the left side is giving me so much anxiety

Thank you for explaining minesweeper in such a clear manner, I always had trouble figuring out what the numbers actually meant.

Never wouldve guessed this was a small channel, amazing work

Love the result. Very smart way at speeding it up, as well.

Dude this is awesome. Subscribed!

Looks great, you've earned a subscriber. Keep it up!

I recommend the game "14 minesweeper variants" it has a built in SAT solver to procedurally generate puzzles that require complicated logic

I like this and I hope you get even more followers. Subscribed.

Awesome quality, keep up the good work

Code Bullet looked at the problem 4 years ago. His solving algorithm is a bit more fancy than yours. If it finds a case where 1 in 3 configurations has a bomb while its neighbors have bombs in 2 in 3 configurations, it will select the 1 in 3 location instead of the 2 in 3 locations reducing the likelihood of bombs going off.

Underrated I can't believe that you don't have 1k subscribers yet

So that's why the 3D pipes screen saver was so popular!

Sometimes, a minesweeper puzzle is actually unsolvable. If your solver finds one of these, I suggest having it cheat a little and give itself a hint, revealing a clear tile.

241 subscribers? This video was amazing keep it going bro I believe in you

I wanted to create my own puzzle solver engine for a puzzle game that I dont know the name for. its a grid of pipes and ends. all you can do is rotate the blocks in place to move the connections around. everything must connect to the center tile. I learned quickly that the edges should be solved first since all edge tiles cannot have a connection. so any "T" shape must always face inwards. allowing you to "lock" the tile and mark it as solved. I thought it would be a fun task to make a program that would solve the puzzles

Wow, I had the same idea over a year ago but had to abandon it due to horrible code base. Thanks for making this!

What if you tried alternative tile shapes? Like hexagons or triangles? Maybe even alternative spacial geometries like the surface of a torus or a cube? That'd be cool, and probably a fun challenge to figure out lol

For the logic around the 02:40 mark, it would make sense to pause the normal "iterating over a bunch of tiles to look for trivial solves" and to look specifically around the two cells forming an XOR and examine the surrounding cells with that characteristic in mind. That way you could blast through most cells very quickly and only apply advanced logic when needed.

Years ago, I made my sister an automatic solitaire game, that played automatically. It was good, too. She didn't think it was as funny as I did.

Very cool! The original minesweeper had a maybe flag that I used extensively :⁠-⁠)

I played this obsessively when it first came out - so much so that I "used" it to think about other things, keeping the overactive part of my brain occupied. When it became too "easy," I stopped dropping bomb markers so that I had to rely just on the number patterns. It was especially satisfying to finish a board with no bomb markers in view. He left out that the original game displays a count of the total bomb markers remaining, which was one of the parameters for resetting the board. Getting to only 5 bombs left and 20 or 30 unmarked squares made it easier to guess safely. The original game "cheated" in that the very first square was ALWAYS NOT a bomb - which means the pattern couldn't be set until AFTER the players first move. (A few times, the first click revealed an 8, though, making the next click VERY risky! 😁

Bruh how tf does this only have 94 views. Very impressive stuff keep it up 👍

I did the same thing with my Sudoku solver (which will also do puzzles that overlap/have common squares). I don't do the exhaustive search from the ComputerPhile videos (he did his in Python as well), but rather tried to articulate "what do I do when I solve?" into algorithmic checks of escalating complexity (Level 1, look for the obvious. Don't find any? Level 2, look for slightly harder to find things. Only make a few pencil marks? Level 3... etc. Every time an actual guess is made I drop all the way back to Level 1 complexity and start over until I've exhausted all techniques without any guesses or new pencil marks). It doesn't know advanced techniques because... well... neither do I hehehe nor could I articulate those techniques I don't think, but it does pretty well.

Liked for your sharing the code.

This would make for a fantastic After Dark screensaver.

yeah, the part at 2:30 is what I call the law of 1-2-1. Wherever you see that pattern, or anything that reduces to that pattern, then the flags go adjacent to the 1's and the center is cleared and depending on how you arrived at 1-2-1, you may also clear whatever is on the other side of those 1's.

Okay, different angle on solving the maybe situation in plaintext: Until a new number is revealed, treat each cell as a rolling percentage of likelihood of a bomb. The first 2 knows that those two cells each have a 50% chance of being a bomb, so for the value of "likelihood of a bomb" set a value of 0.5, and move onto the next digit down the line. The next two adds up the values of all cells it sees to get the total likelihood of bombs placed, reading flagged bombs as a 1, and numbers, spaced, and unclicked cells as 0. It sees that the likely number of bombs placed is 1 (if it were somehow 1.25 or something, it would truncate at the ones place, and do the math from there) so it knows that in any 0 likelihood unclicked cells it sees, they must hold *number in cell* minus *bombs placed value* bombs. in this case there's only one 0 probability unclicked cell the number sees, so it puts all 1 bomb into that cells value. Then the logic decided if flags must be placed. Any cell who's probability of a bomb is one or greater, gets flagged as a bomb, so the cell directly in front of the 1 gets flagged. The logic then reverses the order of thinking and says "does this cause any number I'm touching to be maxed out?" If it finds more than one, that doesn't matter, but if it doesn't find any, the logic would have then continued down the line of numbers, erasing memory of values holding probability of a bomb, continuing to try and find a maxed out number. When it maxes out a number, that number flips all remaining unclicked cells to safe, then starts the process all over again. I'm the kind of guy who feels like he's good at talking out programming logic, but no good at actually writing it. I have no idea if that's a valuable skill, and I'm just hopeful that I made some level of sense, and all respect for having the knowledge to make that programming happen. I hope the idea helps get the juices flowing, should you want to try to make your code leaner.

Chopin is a nice touch

very well made video!

Oh snap. I might have to code this in c#. Years ago I got bored and coded reversi client server over tcp. That was fine and similar logic with checking neighbors.

I have a suggestion for you, I'd implement it but I only have my work laptop atm and can't download anything executable. Make an infinite scrolling mineweeper solver, prioritise solving on the left of the screen, and when the whole let edge is solved move one over, repeat infinitely.

really cool actually!

Only 12 subs? This channel is super underrated I will be the thirteenth sub

This already looks better than may of those crappy screensavers in Linux's Xscreensaver :q

I made a minesweeper screensaver a couple of years ago. I made mine scroll infinitely so it looks like an infinite minefield. I also cheated a little and let it choose a random safe tile to clear when it was stuck.

This made me want to play minesweeper

You can make it 100% solvable since you create the field and therefore know where the mines are. When your solver is stuck, "guess" any empty tile, or one that is next to an empty/number tile.

Amazing! Are there any situations where two hypotheses are required? like, assuming one tile is maybe cleared reveals some data but not enough, so you need to assume/guess about another tile inside the first assumption?

At least at the mine density I always play for custom games (= ratio for expert game, or one mine per 4.8 tiles) I almost always end up with multiple 50/50 guesses unsolvable by logic so it becomes having to win (N) coin tosses.

"hi, i made a" is its own genre now. what a world.

Now all we need is a dark version

Fun fact, you actually only need the global flag on a variable if you want to set a variable from an outer scope. Python actually propagates outwards and finds the first scope with a matching reference. You only use global when you want to write to the highest form of a variable, usually a global. It's actually frowned upon to use globals usually, as it makes it difficult to see where a variable has been changed at, but that doesn't mean it has no use, obviously it's there for a reason. Anyways, thought I'd let you know about that, cheers!

3:56 Um, why this image is so wrong...

At 2:23, there's a 3 along the left side that can't have 3 mines next to it. There are only two places to put mines next to it. The 2 along the left side isn't possible either.

Get the puzzle generator to talk directly with the puzzle solver and cut out that thing in the middle

you should make a 10-hour video of the robot solving it. maybe it could have music or something?

You can use something like convolutions to solve this. Store a probability value for each unrevealed tile neighboring a revealed tile. For all revealed, numbered tiles, offset the probability values of their unrevealed neighbors so they add up to the value they show. When an unrevealed tile is neighbor to N revealed numbered tiles, average their offsets computed by their neighbors. Repeat this until convergence. Reveal the tile with the lowest probability. Repeat until all tiles with a probability less than 0.5 are revealed.

Man you need more than 30 subs

Please put this on a webpage so I can make it my desktop wallpaper!

wayy too underrated

by assigning entropy to each square is best method to solve minesweeper. like reveled square has entropy 0. distribute value of square to surrounding not reveled squares. like if there are 2 not reveled squares around '1' then give them 1/2 & 1/2 entropy. squares having entropy 1 is bomb.

It would be nice if you would design an always beatable minesweeper. I’ve heard they’ve been programmed but never seen one.

I feel called out by the intro lmao

It would be interesting to set this up to scroll infinitely

Counting the bombs also helps at dead-ends

2:23 That's an odd number 3... On the edge, with 4 of 6 spaces revealed....

all things considered this is really cool! though it defeats the main purpose of a screensaver, since there's many static elements that would burn into your screen

Made my day :)

Would there be any possible way you can make a dark mode version of this, and upload it Wallpaper Engine? I would like to have something like this as a Wallpaper Engine application.

Very well done!!! Why is it that our type of channels have so few subscribers? Is it that it's too niche? Too nerdy? Too complex?

Wow I never considered an algorithm that solves via proof by contradiction. How quaint

Do a 100x100 grid with 10,000 bombs. Then do another with 9,999 bombs.

Nice.

I have been toying with the idea to make a solver, but the purpose is to create 100% solvable games, where if you play fully by logic you do not need random to finish it. One thing I see is that the opening move needs to be "pre played", ie one does not choose the first click.

nice

The board at 3:39 is not possible IMO. The lower number 3 on the left being on the edge of the board implies that the two tiles to the right of it are explosives, as well as the tile directly beneath it. So the number 2 gets satisfied by the two tiles that were to the right of that number 3 - but then there has to be an explosive directly below the number 1 to the right of that number 3, which then means there are three explosives on the perimeter of that number 2. That number 2 should therefore be a 3. Or there must be more tiles to the left of that lower number 3 (it can’t be on the edge of the board).

Luke, would you mind telling what happens if the algorithm stumbles on a 50/50 guess? I remember getting some 50/50 when I used to play the game.

You should try to write a solver for SkiaLogic type puzzles.

It’s proof by contradiction!

Only 800 subs? Come on! Take mine!

Nice

Maybe a Tetris solver will please your brain similarly

Did you model this with a Markov Model and/or a Bayes Net?

remember me when you get big :)

infinite scrolling mine sweeper, it tries to solve the top-most cells it can, it can't reveal anything about the bottom row of cells, every time a top row is cleared shift everything up and add random bombs, maybe keeping the top 3rd ish solved and slowly scrolling and adjusting the speed of the solver to keep about 2/3rds +/-5% of the cells solved at any given time

neat!

3:40 the top 3 is wrong, there are only two tiles in its vicinity and no flags

What is the time complexity of the solver?

At 2:45, that board is not possible. Reasons being is that 1. Because we can clearly see that the two sides of the board both show borders, therefore the top 3 cannot be possible as there are only 2 mines surrounding it in an 3x3 area (excluding the border ones). Thus the top 3 cannot be possible. 2. The bottom 3 is possible as it shows that there are 3 mines surrounding it and that there are only 3 available tiles, however using that logic, the 2 beside it will also be clear, leaving the 2122 pattern being like: 3 * 2 1 2 2 * * * x x * * 2 As you can see the top right most 2 cannot be possible as there are 3 mines beside it based on the boards logic

Wait this whole time its been "bombs in the surrouding 8 squares" and not "spaces away that the nearest bomb is"? Wow I've been playing wrong

This is pretty nice!.. So, can i use it as a screensaver?

How do you only have 1k subscribers

This is cool as fuck

Is their any way to turn this into an actual screen saver?

What happened at 5:11? It looks like the solver hit a bomb next to a two tile? Why would it be guessing when the 2 is only touching 2 squares?

wallpaper engine when?

2:23 top left most three shouldnt be there, since only two spaces around it are undiscovered and none of the discovered spaces are bombs.

I don't mean to nitpick, but the board at 5:11 had a tile that could have been confidently cleared, but the game guessed (and blew up.) In the top right, there is a 4 with an unflagged bomb below it. 2 tiles to the right of that bomb could have been cleared, given what was known at the time. My Python isn't very strong, so I doubt I can realistically dig up why the code didn't find that.

the highlighted tiles during the explanation are pretty hard to see

Someone should port this to make it a Android live wallpaper or something

I suggest you maybe try implementing this since it might be fun (might take a while though): Loop through all revealed tiles to find the numbered ones. The number on these tiles are the sum of mines of their neighbouring unopened tiles, so you can describe *all* the information of a minesweeper game in a big (augmented) matrix constructed from these sums (think of the total mine count as an opened numbered tile neighbouring every other tile). Group all the columns that are exactly similar into one since tiles representing these columns have equal probability of containing mines, creating a smaller matrix. Change the matrix into reduced row echelon form, then remove all rows with only 0s at the bottom. Now every single solution in this matrix represents a valid mine distribution within groups (i.e., describing that there's x mines within this group of cells) and the solution space is the number of rows minus number of columns of that matrix. Solve for constraints of every group (e.g., this group can only contain between x and y mines in it) using the fact that: every group must have at least 0 mines, at most, the number of tiles in that group, and using the matrix. Iterate through the solutions space (every solution is dictated by only a few "free" variables), checking if they're valid or not using the constraints. Then you can toy around more with the combinatorics to work out the number of solutions where one set tile in each equiprobable group is a mine. Assuming that every configuration of mine is equally probable to occur, you get the *exact* probability of each tile being a mine or not.