I lost interest because I realised that the constraints are just fancy tracings on top of the core number deduction game which is itself not very complex.
The magic was destroyed, but it's not as if I would have been able to create anything more than a superficial illusion anyway.
I repurposed this LP example and it still solved the hardest puzzles in 100ms, while the regular puzzles were somewhat faster.
That's pretty impressive to me, since other approaches were slower on the hard puzzles and I hardly had to do anything.
It's probably possible for a backtracking heuristic to be a bit better, but could be annoying to figure out on my own.
Now the question is whether this LP-based Sudoku solver is sufficient for generating puzzles.
It needs to be run twice every time it's used with an extra constraint negating the first solution, to detect if there's more than one solution.
Since the solver will be run many times, it's probably still going to be too slow to be useful in the browser.
However I want to try and make a local generator anyway, it's still usable like this.
It would be nice if the user can choose which types of modifier they want, but it's not necessary.
Every Sudoku strategy generalises to its variants, because they all deductively eliminate penciled-in possibilities.
Perhaps if we express the puzzle in terms of boolean logic it can be solved more easily using some existing mathematical technique like linear programming.
Or this could just be another way of expressing the same thing, and the constraints fundamentally aren't easy to solve.
Nevermind, people already do this. Pretty cool.
My intuition is that whatever linear programming algorithm you use has to be faster than any other backtracking search with whatever advanced strategies, since Sudoku can be perfectly represented in this way without information loss.
According to a friendly LLM, the regular search is probably faster since it can use a heuristic to guess squares with the fewest options (which was what I was already doing).
It would be really nice to be able to express all constraints mathematically, so I'll at least try and see if it can solve the hard puzzle that my semi-naive solver took too long to solve.
This is fun. It's like chess programming except I'm actually able to understand it.
I guess it was just a matter of implementing it, no real issues with performance once I did the basic obvious optimisation.
That's more of an Advent of Code problem (derogatory) than a Project Euler problem.
It was fairly fast, but not fast enough to be run many times during puzzle generation without taking like a minute to generate a puzzle.
I might have to settle for locally pregenerated puzzles, which I think is the usual tradeoff everyone else makes.
Sudoku.js made another tradeoff, but a horrible one that completely compromises the reliability of the algorithm, seems pointless.
Or maybe I could just improve the algorithm, I could do more, for instance I didn't use any complex Sudoku strategies before I started guessing.
Someone in the forum posted some more difficult puzzles, and my algorithm is way slower on those.
But it seems like there's still room to improve.
I also didn't use much memory because I just wanted to get something that worked even if it was slow.
Since the Sudoku variant puzzles I mentioned below (Kakuro, Killer Sudoku, Str8ts) all have similarities, perhaps it's possible to make a generic puzzle that can use any of the constraints from these variants.
Pencil markings are used to solve all of them, so it would probably be easy to design a UI.
Kakuro has the greatest divergence by removing the restriction of no repeats in a row or column, instead restricting that no repeats can exist within sums.
I would most likely abolish the Kakuro rule since the difference is subtle.
The no repeats in row/column rule feels best represented as a static game rule affecting the entire board and not specific tiles, but this could perhaps even be abstracted away.
Str8ts and Kakuro also both remove the 3x3 squares from Sudoku, but I want the 3x3 squares to be optional.
I also found a bunch of other variants here and here, although most of them would not work well here.
A composite puzzle is inherently inelegant, but I hope that it could lead to great emergent complexity, and the possibility of unique visually patterned puzzles.
Modifiers would apply effects to one or more squares in a contiguous region.
The region doesn't have to be rectangular, e.g. in Killer Sudoku they have strange shapes, but it has to be contiguous because otherwise it's extremely inelegant.
Effects could include:
- No repeats
- Consecutive (Str8ts)
- Non-consecutive (not the opposite of consecutive)
- Sum to given value (Killer Sudoku, Kakuro), inelegant
- Even/odd
- Obstructions (Str8ts, Kakuro), may or may not hold a value
- Multiple numbers fit in a single square
- One number must be double the other (Kropki, 2-size region only)
Since the region is contiguous, requiring all squares in the region to be the same digit is impossible unless we relax the row/column constraint, which is a possibility.
This sudoku.js library is awful (and at least someone noticed), I don't know why it has so many stars.
It seems possible to create my own algorithm, although generating a sudoku puzzle from scratch with a single solution is NP-hard.
A lot of answers online suggest permuting an existing puzzle, which is also a horrible recommendation.
The algorithm used to find solutions is a simple backtracking algorithm that could be easily generalised to include other modifiers, because the rules only apply when it checks to see if the game is solved, something that can easily be computed for any custom rule.
In order to make the solver algorithm efficient, ruling out possibilities using pencil markings would probably be sufficient and would also generalise across any custom rule.
Once you have the solver algorithm, you just have to remove numbers from a solvable puzzle (initially with all numbers filled in) until you reach a state with only one solution.
Since you wouldn't always arrive at a puzzle with one solution, backtracking should be used here as well.
This could be really slow, but I can't tell until I try.
I want it to run in the browser, the speed could also depend on the number and types of modifiers.
I never did Project Euler #96 (probably because it looks inelegant compared to most PE problems), maybe I'll learn more about the puzzle if I try that first.
Actually, the fact that only 50 solutions are required is menacing.
PE problems almost always have performance challenges, and it's 25% difficulty, so maybe the simple backtracking algorithm won't be sufficient.
Killer Sudoku and Kakuro are inelegant because they require summation, which is tedious.
Sudoku and Str8ts are advantaged by having constraints that demand less calculation, although Sudoku generally does not have enough emergent complexity to be a good puzzle.
#200 I could link together two colours to create colours for one of those optical illusion images where the shadowed colour is the same as a different lit colour.
It looks too hard to make a chess engine on my own.
I found Caissa which is written in C++ and has an MIT license so I can use it for anything.
Just running an engine and changing some parameters with the UCI protocol probably isn't sufficient for my needs.
Thankfully, Caissa builds with MinGW make even though the instructions recommend creating a Visual Studio project.
I tested it with Arena which is a GUI for chess engines that implement the UCI protocol, and it seems to work properly.
Now I just need to see how realistic it is to modify this engine to suit my needs.
It seems like some UCI commands to control the engine measure time in seconds, which isn't what I want.
I spent a while reading the code but it's way too complicated, it says it uses 7 different search algorithms.
It doesn't seem to use NNUE but it uses a neural network of some kind.
Search.cpp seems to be the important part, but I don't have a clue what's going on.
It's possible that all good chess engines might be like this.
I found a very simple chess engine written in only 111 lines of python called sunfish, but I don't think it's strong enough since many human players can beat it.
Not sure how to find an engine in the middle ground between the two.
#213 - I just realised that the animation speed disparities are probably related to the time taken to perform Dijkstra's algorithm differing for each character.
Not sure if that's only an Emscripten thing or if it's in the desktop version too.
I checked and it's on desktop as well.
I think the way to show the player's abilities influencing the AI would be to start off showing the best moves, then activate each item and update the best move arrows visibly in sequence.
Lichess shows a varying number of best moves, and scales the arrow size depending on the relative weighting so you can see how relatively good the moves are to one another.
I think Balatro succeeded where other indieslop games failed because there is a clear separation between the familiar basic mechanics and the elaborate scoring system.
Other attempts to put a spin on classic games mutate the core gameplay too much and create something that is no longer familiar, and probably damage aspects of the original game that were good in the process.
Stone cards are Balatro's worst feature because they don't align with the base poker hands.
Games based on chess tend to introduce new "innovative" elements that are highly disruptive.
Often all that remains is "pieces moving in a familiar pattern on a board", and the accumulative strategic focus of real chess is trashed.
I even consider Chess960 to be bad. I think the familiarity of the starting position is important.
Crazyhouse is fun albeit.
I wonder if it would be possible to build a scoring layer on top of chess without damaging the core gameplay.
The following elements should be sacred:
- Turn based
- Starting position and pieces
- 8x8 board
- Ruleset
If so much of the game must remain unchanged, what is there left to vary and innovate upon?
Balatro still allowed the deck to change from the regular 52 cards, creating new hands like 5 of a kind, which sacrifices some of Poker's integrity.
However, I think chess is different to poker and requires the elements listed above.
The element that can be varied is the enemy AI.
Balatro creates "chips", "mult", "hand level" out of thin air, so I could similarly create new elements that form a system that allows some degree of influence over opponent AI.
Playing against the strongest Stockfish is really unfair and impossible for a human to win, but what if you could dynamically handicap it as the game progresses?
Having a complex handicap system seems like it would be more interesting than playing against weaker levels of a chess AI, which isn't very fun.
Chess skill varies wildly, but learning to use the new handicap system would be a learning curve for all players.
The ascension system could simply be the AI's depth scaling up.
Relic chess interaction ideas (it will also interact with currency and relic systems):
Relics that trigger upon castling, relics that incentivise preventing enemy castling (hard).
Trigger upon capture.
Trigger upon pawn promotion (very hard).
Trigger upon en passant (very hard).
Trigger upon check.
Trigger upon equal/lesser/greater material score (e.g. pawn = 1, knight = 3, etc).
Trigger upon X piece moved/captured.
Alter AI depth.
Alter AI to prefer/dislike moving certain piece types.
Alter AI to over/undervalue specific pieces (NNUE evaluation). Permanent.
Alter AI to over/undervalue control of specific squares. Permanent.
Entirely block moving to a specific square for one turn (expensive, might have to limit this so it doesn't assist checkmates).
Entirely block a specific piece from moving for one turn (see above).
Alter other AI stats related to aggressiveness, liking to trade, or other variables I'm unaware of?
Clearly, "trigger" should be a keyword mechanic.
I would limit relics that give information to the player, and instead focus on depriving the AI of information.
Maybe the player could have a relic that shows them the computer evaluation, or shows the enemy's threat move, but showing the player what move they should make is uninteractive.
The biggest challenge in implementing this would be creating a custom chess engine that I can dynamically influence, but maybe there are enough resources available that this would not be prohibitive. I don't think it would be possible to just use Stockfish, but I want something of comparable strength that beats all humans.
I know a lot of people make their own chess engines, but I don't know how difficult it is to approach Stockfish or if it's easy for a hobbyist to get there nowadays.
NNUEs are pretty simple, so it would be nice to make use of that.
To make the gameplay robust, I guess it the game speed would decrease on lower end computers, since cutting the AI off based on a time limit would be unfair, but hopefully I could get it fast enough that I don't have to worry about this too much?
Another issue is that it would be difficult for the player to observe how the player's relics influence the AI, and they may feel cheated if the AI makes a move they attempted to guard against.
Visual representation of certain pieces/squares being buffed/debuffed could help here, as well as showing top moves before the move plays out.
Could also visualise NNUE evaluations.
I'd also have visible stats for depth and amount of pruning.
I'm not sure if it would be good to always show the overall AI evaluation score, or only if you purchase a relic. I'm thinking always show it, because that means it can be used to trigger other relics.
It will also be difficult to balance, but at least it's safe to assume that any alteration to the AI should weaken it to some degree, even if it's slight.
A lot of iteration on concepts would be required to make this a reality.
I really love forum mafia, but it requires too much effort.
I have a tendency to overfocus on things to the point where it causes me physical harm.
A forum mafia game is a very challenging social puzzle that basically has no upper limit to the amount of effort you can put into it, so it can be rewarding if you try very hard.
Even if I tell myself to take something less seriously, I get swept up by the allure of an unsolved problem and trapped in focus.
Another aspect is that the reward gradient is fairly constant, it always feels like you're making progress after every post (and this isn't really illusory, you really do need to immerse yourself gradually to develop a sense of the gamestate).
Other sorts of puzzles often have a few moments of sudden insight interrupting blind exploratory stumbling.
You never feel stumped in a forum mafia game because there's always more you can do.
It's easier to take a break from a puzzle when the next big insight doesn't feel reachable for now, but if there's lots of small insights, it never feels like the right time to stop.
I'm reading Reverend Insanity, which is comically evil Chinese shonen.
It's more than twice as long as Worm, so I don't know if I'll end up finishing it.
It was banned in China, I think because it's subtly subversive to the regime by cynically deconstructing how social power structures are maintained.
I uploaded this game a while ago, but since it has been dead and coped for several months now I might as well post about it.
The controls are LMB to move, Z to undo, X to switch, and C to separate.
The core concepts of the game are Dijkstra's algorithm and the pair up mechanic from Fire Emblem Fates.
I had an idea for a simple story but I never made more than this demo room which has nothing to do aside from move around, and I didn't really add any puzzle mechanics like doors or flags.
I'm not sure if I would be able to design elegant puzzle mechanics or not, but I think full path reachability being shown is quite interesting and there's probably something I could do.
----
If I wanted to continue with this project, I would start from scratch and design the action stack and undo feature first, because I had a horrible time with that.
The animations are messed up and there are still bugs with the undo feature.
I would also not use SDL, because it's not enjoyable to use, although it's not that bad for 2D.
I at least learned how to build an SDL application for WebAssembly and I was impressed by how easy that was, but I'm sure you can do that with raylib and probably Godot or whatever else people use anyway.
I think it's common, for those who seek meaning in art, to find it comforting that art can persist for a long time.
Because the corpse of the artwork seems permanent, the whole experience is felt to be also, even though it is not.
For this reason I think it is important for those who seek to create meaningful art to focus on the aspects of artistic creation that have timeless value, and avoid drawing excessively from other works that require social context to understand.
The process of artistic creation has to some extent a temporal social aspect that is similar to the process of consumption that creation appears to transcend.
I can't understand the mindset of someone who dedicates themselves to preserving media they find nostalgic.
There is no opportunity for growth, what existed once is all there is.
They have boxed themselves into stagnation, isolation, death.
I can't decouple nostalgia from sorrow.
I used to save a lot of webpages and files on my computer, but they put me in a terrible mood whenever I looked over them.
I don't want to remember anything.
The future has gotten worse, but I have never found hiding in the past appealing.
I would prefer to imagine a positive vision for the future that solves modern problems and also takes advantage of modern boons.
----
The spirit of old media has vacated, leaving only the perfectly preserved corpse.
Multiplayer games are abandoned.
Your friends will not understand or care about your experiences.
No new media will branch off from this point, and if an attempt is made it will wither on the vine.
We have long denied the inherent social aspect of even kinds of leisure that seem asocial.
There's an exception in "classic media" like Dostoyevsky or Runescape, which seems to endure beyond its expected lifespan.
Revival of old trends or styles to bait nostalgia always feels insincere because the spirit of that time has already passed.
Two creative writing techniques I observed from reading:
Objective writing:
Choose a goal of a particular scene or event in the story, and challenge yourself to write convincing plot events that allow the story to reach the target state.
For instance in Worm, it seems like the author Wildbow is often allowing a threat to loom against the protagonists, and then challenging himself to help the protagonists overcome the threat.
Since it was released serially, he wouldn't have a precise plan of how the story will unfold, but it seems like this could be useful for traditional writing too.
I also got this impression from how Wildbow claimed he used randomisation to determine the fate of characters in an early part of the story, and it was even possible for the protagonist to die and the story's focus to shift to a new character (this contains spoilers for Worm, if you aren't sure then don't click this).
It seems like he treated the process of building a story as a puzzle to solve, rather than a gradual construction.
Imperfect coincidence:
Intentionally choosing story elements that almost cohere but have a slight difference from one another can be used to create mystery.
It can also often involve two or more objects or concepts that seem identical at first glance but have a subtle difference.
Often the reader might notice the imperfection but write it off as a plot hole, negligible detail, or red herring.
I think it's positive if the reader is thinking deeply about a mystery story in this way.
Of course, it could be used as a subtle red herring as well.
There are many examples of this in the When They Cry visual novels.
song15.mp3:
song17.mp3:
song18.mp3:
I believe that advertising is intended to influence investors more than consumers.
Added title/image/description meta tags to the individual post page.
It should use the first image if there are multiple.
Here's some screenshots from another dead cope project I am working on using raylib, might build it for WebAssembly and put it here eventually if I haven't given up by then:
A naive approach to creating a mesh from a heightmap can lead to inconsistent results.
If we consider the 4 adjacent points around one pixel of the heightmap, they cannot be formed into a quad because they will not necessarily fit on the same plane.
We need to split the quad into two triangles, and there are two different ways to do this, by splitting along either of the two diagonals:
If we always pick the same diagonal, this creates inconsistent looking results on smaller heightmaps.
I think OpenGL works this way if you use GL_QUADS. There is no "correct" diagonal to pick.
In order to resolve this, you could place an extra vertex at the center of the quad, and then draw 4 triangles instead of 2.
Delaunay triangulation is another potential solution to this problem by getting rid of the pixel grid entirely.
Access to good real-time financial data from big name companies can cost tens of thousands of dollars in subscription fees per month.
In today's economy financial data is not practically that expensive to provide.
The price is kept high on purpose to appease big traders who want to ensure access to market data is artificially limited.
That this is not more commonly recognised as a scam is a communication failure on the part of humanity.
If information were a more directly visible input with the same priority as money, for example, I think this would be more widely understood.
There aren't good words for this, but "emergent conspiracy" comes to mind.
The circumstance is likely unplanned and its mechanisms are informally coordinated, but it nevertheless enriches a minority at the expense of the majority, like a traditional conspiracy might.
Instances of asset speculation ("distributed pyramid schemes") are similar.
You can't directly add opacity to a named colour so I will preserve the colour switching script and allow for shades of colour by using an absolutely positioned ::before pseudoelement as a background.
I noticed that Verdana 14px looks good, while Verdana 18px does not.
Consolas 18px looks a bit better than Consolas 14px.
Consolas is similar but simply better than Courier New so I will prefer it over "monospace" although it seems Consolas is not widely supported.
Oh, monospace is Consolas on my system already, but the tilde stylesheet I used specifies Courier New. All the more reason to be rid of it...
#198 Added the opacity tool.
I chose never to vary the opacity and recompute the other output colour if one changes, which simplified things a lot.
Another limitation of this is that an output colour sometimes cannot be reached at all if it differs too much from the input colour unless the opacity increases, and I chose never to vary it, so sometimes the results will be incorrect.