I posted this question over at CodeReview, but I came to realize that it's not so much a Haskell question as it is an algorithm question.
The Haskell code can be found on my github repo but I think the code isn't as important as the general concept.
Basically the program figures out the optimal first set of moves in a game of Kalaha (the Swedish variation). Only the first "turn" is considered, so we assume that you get to start and that anything from the point of the opponent getting to move is not computed.
Kalaha board http://www.graf-web.at/mwm/kalaha.jpg
The board starts off with empty stores and a equal amount of marbles in each pot.
You start your turn by choosing a non-empty pot on your side, pick all of the marbles up from that pot and then move around the board by dropping one marble when passing a pot. If your last marble lands in the store, you get another turn. If you land in a non-empty, non-store pot, you pick up all of the contents of that pot and continue. Finally, if you land in an empty pot, the turn is passed to the opponent.
So far I've solved this by picking all possible paths and then sorting them according to the amount of marbles in the store by the end. A path would mean to start from one of your pots, do all the necessary pick-up-and-move-around and see if you either land in a store or in an empty pot. If you land in the store you get to continue, and now there are as many new branches as there are non-empty pots on your side.
The problem lies in the fact that if you start with five marbles in the pots, it's already quite a few paths. Jump up to six and ghci runs out of memory.
The reason I can't figure out how to make this less expensive is because I think that each path is necessary during computation. While I'll only need at most three paths (the best ones) out of the thousands (or millions) generated, the rest need to be run through to see if they're actually better than the previous ones.
If one is longer (generally better) then that's good but costly. If it's shorter than any previous path, then program still had to compute that path to find that out.
Is there any way around this, or is computing all the paths necessary by definition?
Just try them all out in order, recording your sequences of moves as sequences of numbers 1 through 6 (representing the pot that you pick the marbles from), each time replaying the whole sequence from the start. Keep and update just the three winners, plus the very last sequence of moves so you'd know what to try next. If there's no next legal moves, go back one notch.
It's going to be prohibitively slow perhaps, but will use very little memory. You don't store resulting positions, only numbers of pots picked from, and replay the sequence anew each time from the starting position, altering the very last move (instead of 2, next try 3, 4 etc.; if no more legal moves, backtrack one level). Or maybe store positions just for the very last sequence of moves tried, for easier backtracking.
It's a typical space for speed trade-of then.
You could try parallellizing using this parallel version of map:
parMap :: (a -> b) -> [a] -> Eval [b]
parMap f xs = map f xs `using` parList rseq
Then you will spark off a new thread for each choise in your branch.
if you use as parMap pathFunction kalahaPots as your recursion, it will spark a lot of threads, but it might be faster, you could chunk it but i'm not that good of a parallell haskeller.
Related
I basically have an algorithm, but it is really slow. Since my algorithm/problem is so simple, I expect, that this may exist somewhere (in fast) and there might also be a name for this. And before I start developing a faster version of my algorithm, I first try to ask here (I don't want to reinvent things).
The problem is simple: I have a time series from an experiment, which is quite large (~5 GB). The thing is, that most of the data points are placed on a line, e.g.
(t=0.0, y=0.0), ... , (t=1.0, y=0.5), ... , (t=2.0, y=1.0)
This could obviously be simplified by interpolating the first and the last point with a straight line. In principle, I can test, if the points between an interval can be approximated by a straight line, within some tolerance (I don't need lossless compression) and throw away the points in between.
My current algorithm works as follows:
I have points within an interval [a,b] and I create a linear interpolation between the first and the last point (let's call this interpolation f).
Then, I compute the error Abs(f(t) - y) at each time series point and select the point, with the largest error (let's cal this point tmax).
I split the interval [a,b] -> [a, tmax], [tmax, b]
Repeat my algorithm on the sub intervals, until a tolerance is reached, or the interval contains only one or 2 points. Return the interval boundaries.
This algorithm works surprisingly well in approximating a signal, but it is really slow and as already said, I believe that there exist already something, which does the same thing or solves my problem.
Thanks for the help, if anything is unclear, don't hesitate to ask.
It looks like you want the Swinging Door compression algorithm. It basically works by using the mental image of a pair of doors to quickly absorb points into a range that can be approximated by a single straight line. It shows up a lot for processing time series in industrial automation. Which is a domain where people wind up collecting a lot of data, very quickly, and needing to summarize it on the fly before doing other calculations.
I won't explain it because there are plenty of good explanations out there, with source code. Here a links to a couple.
Swinging Door in PostgreSQL
Swinging Door in Python
I'm trying to implement a MCTS algorithm for the AI of a small game. The game is a rpg-simulation. The AI should decides what moves to play in battle. It's a turn base battle (FF6-7 style). There is no movement involved.
I won't go into details but we can safely assume that we know with certainty what move will chose the player in any given situation when it is its turn to play.
Games end-up when one party has no unit alive (4v4). It can take any number of turn (may also never end). There is a lot of RNG element in the damage computation & skill processing (attacks can hit/miss, crit or not, there is a lots of procs going on that can "proc" or not, buffs can have % value to happens ect...).
Units have around 6 skills each to give an idea of the branching factor.
I've build-up a preliminary version of the MCTS that gives poor results for now. I'm having trouble with a few things :
One of my main issue is how to handle the non-deterministic states of my moves. I've read a few papers about this but I'm still in the dark.
Some suggest determinizing the game information and run a MCTS tree on that, repeat the process N times to cover a broad range of possible game states and use that information to take your final decision. In the end, it does multiply by a huge factor our computing time since we have to compute N times a MCTS tree instead of one. I cannot rely on that since over the course of a fight I've got thousands of RNG element : 2^1000 MCTS tree to compute where i already struggle with one is not an option :)
I had the idea of adding X children for the same move but it does not seems to be leading to a good answer either. It smooth the RNG curve a bit but can shift it in the opposite direction if the value of X is too big/small compared to the percentage of a particular RNG. And since I got multiple RNG par move (hit change, crit chance, percentage to proc something etc...) I cannot find a decent value of X that satisfies every cases. More of a badband-aid than anythign else.
Likewise adding 1 node per RNG tuple {hit or miss ,crit or not,proc1 or not,proc2 or not,etc...} for each move should cover every possible situations but has some heavy drawbacks : with 5 RNG mecanisms only that means 2^5 node to consider for each move, it is way too much to compute. If we manage to create them all, we could assign them a probability ( linked to the probability of each RNG element in the node's tuple) and use that probability during our selection phase. This should work overall but be really hard on the cpu :/
I also cannot "merge" them in one single node since I've got no way of averaging the player/monsters stat's value accuractely based on two different game state and averaging the move's result during the move processing itself is doable but requieres a lot of simplifcation that are a pain to code and will hurt our accuracy really fast anyway.
Do you have any ideas how to approach this problem ?
Some other aspects of the algorithm are eluding me:
I cannot do a full playout untill a end state because A) It would take a lot of my computing time and B) Some battle may never ends (by design). I've got 2 solutions (that i can mix)
- Do a random playout for X turns
- Use an evaluation function to try and score the situation.
Even if I consider only health point to evaluate I'm failing to find a good evaluation function to return a reliable value for a given situation (between 1-4 units for the player and the same for the monsters ; I know their hp current/max value). What bothers me is that the fights can vary greatly in length / disparity of powers. That means that sometimes a 0.01% change in Hp matters (for a long game vs a boss for example) and sometimes it is just insignificant (when the player farm a low lvl zone compared to him).
The disparity of power and Hp variance between fights means that my Biais parameter in the UCB selection process is hard to fix. i'm currently using something very low, like 0.03. Anything > 0.1 and the exploration factor is so high that my tree is constructed depth by depth :/
For now I'm also using a biaised way to choose move during my simulation phase : it select the move that the player would choose in the situation and random ones for the AI, leading to a simulation biaised in favor of the player. I've tried using a pure random one for both, but it seems to give worse results. Do you think having a biaised simulation phase works against the purpose of the alogorithm? I'm inclined to think it would just give a pessimistic view to the AI and would not impact the end result too much. Maybe I'm wrong thought.
Any help is welcome :)
I think this question is way too broad for StackOverflow, but I'll give you some thoughts:
Using stochastic or probability in tree searches is usually called expectimax searches. You can find a good summary and pseudo-code for Expectimax Approximation with Monte-Carlo Tree Search in chapter 4, but I would recommend using a normal minimax tree search with the expectimax extension. There are a few modifications like Star1, Star2 and Star2.5 for a better runtime (similiar to alpha-beta pruning).
It boils down to not only having decision nodes, but also chance nodes. The probability of each possible outcome should be known and the expected value of each node is multiplied with its probability to know its real expected value.
2^5 nodes per move is high, but not impossibly high, especially for low number of moves and a shallow search. Even a 1-3 depth search shoulld give you some results. In my tetris AI, there are ~30 different possible moves to consider and I calculate the result of three following pieces (for each possible) to select my move. This is done in 2 seconds. I'm sure you have much more time for calculation since you're waiting for user input.
If you know what move the player is obvious, shouldn't it also obvious for your AI?
You don't need to consider a single value (hp), you can have several factors that are weighted different to calculate the expected value. If I come back to my tetris AI, there are 7 factors (bumpiness, highest piece, number of holes, ...) that are calculated, weighted and added together. To get the weights, you could use different methods, I used a genetic algorithm to find the combination of weights that resulted in most lines cleared.
I'm trying to implement a model of predator-prey.
It is agent-based model. Every few milliseconds is a new move. On the field there are two types of creatures: predator and prey. The behavior of each of them is given by the following rules:
Prey:
Just moved to an unoccupied cell
Every few steps creates offspring to his old cell
Life expectancy is limited by the number of moves
Predator:
Predator moves to the cell with prey. If such cells are not, in any
free neighboring cell
Same
Same
I have a problem with the choice of prey move.
For example, I have preys in cells 5 and 9.
Each of them can move to cell 6.
How can I resolve this conflict?
Thanks
Use asynchronous updating. Iterate through the prey in random order, having them decide in turn to which cell they should move.
This is a common approach in simulations. It has an additional benefit in that it eliminates limit cycles in the dynamics.
How long does 'moving' take? If you move one, then after the prey has moved, you move the next one, there is no conflict. The prey will simply see the space is already occupied and move elsewhere.
If moving takes time you might say the prey keep an eye on each other and see if some other prey is trying to move somewhere (like people watch cars in traffic). Then you would change the status of the target field to 'reserved for 5' when prey from 5 is trying to move there. Then prey from 9 can see this and decide if they want to collide with 5 (could be intresting :P) or avoid 5.
Depends on game logic. If preys can be on the same cell, so simply use indicator that show preys count. If you are using 2D array for representing current field state you can use such codes:
-1 - predator
n - preys
n >= 0, (n = 0 - cell is empty, n = 1 cell contains 1 prey and so on).
Otherwise (if preys can't appear on the same cell) use turn-based strategy. Save all your preys in array or give number to each prey. In that case preys' moves represents by simple loop (pseudocode):
for each prey in preys
move(prey)
end
where move logic describes algorithm how your prey moves.
Quite a few ways, depending on if you're deciding & moving as two steps or one, etc:
Keep track of each prey's intended moves, and prevent other prey from occupying those.
Check if another prey is already occupying the destination, and do nothing if so.
Remove one of the preys at random if they both try to occupy the same location.
Re-evaluate the move options if the destination is occupied.
There's not really a 'right' way to do it.
See this related question and my answer.
It describes a good collision detection mechanism.
Avoid O(n^2) complexity for collision detection
Regardless of the layout being used for the tiles, is there any good way to divvy out the tiles so that you can guarantee the user that, at the beginning of the game, there exists at least one path to completing the puzzle and winning the game?
Obviously, depending on the user's moves, they can cut themselves off from winning. I just want to be able to always tell the user that the puzzle is winnable if they play well.
If you randomly place tiles at the beginning of the game, it's possible that the user could make a few moves and not be able to do any more. The knowledge that a puzzle is at least solvable should make it more fun to play.
Place all the tiles in reverse (ie layout out the board starting in the middle, working out)
To tease the player further, you could do it visibly but at very high speed.
Play the game in reverse.
Randomly lay out pieces pair by pair, in places where you could slide them into the heap. You'll need a way to know where you're allowed to place pieces in order to end up with a heap that matches some preset pattern, but you'd need that anyway.
I know this is an old question, but I came across this when solving the problem myself. None of the answers here are quite perfect, and several of them have complicated caveats or will break on pathological layouts. Here is my solution:
Solve the board (forward, not backward) with unmarked tiles. Remove two free tiles at a time. Push each pair you remove onto a "matched pair" stack. Often, this is all you need to do.
If you run into a dead end (numFreeTiles == 1), just reset your generator :) I have found I usually don't hit dead ends, and have so far have a max retry count of 3 for the 10-or-so layouts I have tried. Once I hit 8 retries, I give up and just randomly assign the rest of the tiles. This allows me to use the same generator for both setting up the board, and the shuffle feature, even if the player screwed up and made a 100% unsolvable state.
Another solution when you hit a dead end is to back out (pop off the stack, replacing tiles on the board) until you can take a different path. Take a different path by making sure you match pairs that will remove the original blocking tile.
Unfortunately, depending on the board, this may loop forever. If you end up removing a pair that resembles a "no outlet" road, where all subsequent "roads" are a dead end, and there are multiple dead ends, your algorithm will never complete. I don't know if it is possible to design a board where this would be the case, but if so, there is still a solution.
To solve that bigger problem, treat each possible board state as a node in a DAG, with each selected pair being an edge on that graph. Do a random traversal, until you find a leaf node at depth 72. Keep track of your traversal history so that you never repeat a descent.
Since dead ends are more rare than first-try solutions in the layouts I have used, what immediately comes to mind is a hybrid solution. First try to solve it with minimal memory (store selected pairs on your stack). Once you've hit the first dead end, degrade to doing full marking/edge generation when visiting each node (lazy evaluation where possible).
I've done very little study of graph theory, though, so maybe there's a better solution to the DAG random traversal/search problem :)
Edit: You actually could use any of my solutions w/ generating the board in reverse, ala the Oct 13th 2008 post. You still have the same caveats, because you can still end up with dead ends. Generating a board in reverse has more complicated rules, though. E.g, you are guaranteed to fail your setup if you don't start at least SOME of your rows w/ the first piece in the middle, such as in a layout w/ 1 long row. Picking a completely random (legal) first move in a forward-solving generator is more likely to lead to a solvable board.
The only thing I've been able to come up with is to place the tiles down in matching pairs as kind of a reverse Mahjong Solitaire game. So, at any point during the tile placement, the board should look like it's in the middle of a real game (ie no tiles floating 3 layers up above other tiles).
If the tiles are place in matching pairs in a reverse game, it should always result in at least one forward path to solve the game.
I'd love to hear other ideas.
I believe the best answer has already been pushed up: creating a set by solving it "in reverse" - i.e. starting with a blank board, then adding a pair somewhere, add another pair in a solvable position, and so on...
If you a prefer "Big Bang" approach (generating the whole set randomly at the beginning), are a very macho developer or just feel masochistic today, you could represent all the pairs you can take out from the given set and how they depend on each other via a directed graph.
From there, you'd only have to get the transitive closure of that set and determine if there's at least one path from at least one of the initial legal pairs that leads to the desired end (no tile pairs left).
Implementing this solution is left as an exercise to the reader :D
Here are rules i used in my implementation.
When buildingheap, for each fret in a pair separately, find a cells (places), which are:
has all cells at lower levels already filled
place for second fret does not block first, considering if first fret already put onboard
both places are "at edges" of already built heap:
EITHER has at least one neighbour at left or right side
OR it is first fret in a row (all cells at right and left are recursively free)
These rules does not guarantee a build will always successful - it sometimes leave last 2 free cells self-blocking, and build should be retried (or at least last few frets)
In practice, "turtle" built in no more then 6 retries.
Most of existed games seems to restrict putting first ("first on row") frets somewhere in a middle. This come up with more convenient configurations, when there are no frets at edges of very long rows, staying up until last player moves. However, "middle" is different for different configurations.
Good luck :)
P.S.
If you've found algo that build solvable heap in one turn - please let me know.
You have 144 tiles in the game, each of the 144 tiles has a block list..
(top tile on stack has an empty block list)
All valid moves require that their "current__vertical_Block_list" be empty.. this can be a 144x144 matrix so 20k of memory plus a LEFT and RIGHT block list, also 20 k each.
Generate a valid move table from (remaning_tiles) AND ((empty CURRENT VERTICAL BLOCK LIST) and ((empty CURRENT LEFT BLOCK LIST) OR (empty CURRENT RIGHT BLOCK LIST)))
Pick 2 random tiles from the valid move table, record them
Update the (current tables Vert, left and right), record the Tiles removed to a stack
Now we have a list of moves that constitute a valid game. Assign matching tile types to each of the 72 moves.
for challenging games, track when each tile becomes available. find sets that have are (early early early late) and (late late late early) since it's blank, you find 1 EE 1 LL and 2 LE blocks.. of the 2 LE block, find an EARLY that blocks ANY other EARLY that (except rightblocking a left side piece)
Once youve got a valid game play around with the ordering.
Solitaire? Just a guess, but I would assume that your computer would need to beat the game(or close to it) to determine this.
Another option might be to have several preset layouts(that allow winning, mixed in with your current level.
To some degree you could try making sure that one of the 4 tiles is no more than X layers below another X.
Most games I see have the shuffle command for when someone gets stuck.
I would try a mix of things and see what works best.
To experiment, I've (long ago) implemented Conway's Game of Life (and I'm aware of this related question!).
My implementation worked by keeping 2 arrays of booleans, representing the 'last state', and the 'state being updated' (the 2 arrays being swapped at each iteration). While this is reasonably fast, I've often wondered about how to optimize this.
One idea, for example, would be to precompute at iteration N the zones that could be modified at iteration (N+1) (so that if a cell does not belong to such a zone, it won't even be considered for modification at iteration (N+1)). I'm aware that this is very vague, and I never took time to go into the details...
Do you have any ideas (or experience!) of how to go about optimizing (for speed) Game of Life iterations?
I am going to quote my answer from the other question, because the chapters I mention have some very interesting and fine-tuned solutions. Some of the implementation details are in c and/or assembly, yes, but for the most part the algorithms can work in any language:
Chapters 17 and 18 of
Michael Abrash's Graphics
Programmer's Black Book are one of
the most interesting reads I have ever
had. It is a lesson in thinking
outside the box. The whole book is
great really, but the final optimized
solutions to the Game of Life are
incredible bits of programming.
There are some super-fast implementations that (from memory) represent cells of 8 or more adjacent squares as bit patterns and use that as an index into a large array of precalculated values to determine in a single machine instruction if a cell is live or dead.
Check out here:
http://dotat.at/prog/life/life.html
Also XLife:
http://linux.maruhn.com/sec/xlife.html
You should look into Hashlife, the ultimate optimization. It uses the quadtree approach that skinp mentioned.
As mentioned in Arbash's Black Book, one of the most simple and straight forward ways to get a huge speedup is to keep a change list.
Instead of iterating through the entire cell grid each time, keep a copy of all the cells that you change.
This will narrow down the work you have to do on each iteration.
The algorithm itself is inherently parallelizable. Using the same double-buffered method in an unoptimized CUDA kernel, I'm getting around 25ms per generation in a 4096x4096 wrapped world.
what is the most efficient algo mainly depends on the initial state.
if the majority of cells is dead, you could save a lot of CPU time by skipping empty parts and not calculating stuff cell by cell.
im my opinion it can make sense to check for completely dead spaces first, when your initial state is something like "random, but with chance for life lower than 5%."
i would just divide the matrix up into halves and start checking the bigger ones first.
so if you have a field of 10,000 * 10,000, you´d first accumulate the states of the upper left quarter of 5,000 * 5,000.
and if the sum of states is zero in the first quarter, you can ignore this first quarter completely now and check the upper right 5,000 * 5,000 for life next.
if its sum of states is >0, you will now divide up the second quarter into 4 pieces again - and repeat this check for life for each of these subspaces.
you could go down to subframes of 8*8 or 10*10 (not sure what makes the most sense here) now.
whenever you find life, you mark these subspaces as "has life".
only spaces which "have life" need to be divided into smaller subspaces - the empty ones can be skipped.
when you are finished assigning the "has life" attribute to all possible subspaces, you end up with a list of subspaces which you now simply extend by +1 to each direction - with empty cells - and perform the regular (or modified) game of life rules to them.
you might think that dividn up a 10,000*10,000 spae into subspaces of 8*8 is a lot os tasks - but accumulating their states values is in fact much, much less computing work than performing the GoL algo to each cell plus their 8 neighbours plus comparing the number and storing the new state for the net iteration somewhere...
but like i said above, for a random init state with 30% population this wont make much sense, as there will be not many completely dead 8*8 subspaces to find (leave alone dead 256*256 subpaces)
and of course, the way of perfect optimisation will last but not least depend on your language.
-110
Two ideas:
(1) Many configurations are mostly empty space. Keep a linked list (not necessarily in order, that would take more time) of the live cells, and during an update, only update around the live cells (this is similar to your vague suggestion, OysterD :)
(2) Keep an extra array which stores the # of live cells in each row of 3 positions (left-center-right). Now when you compute the new dead/live value of a cell, you need only 4 read operations (top/bottom rows and the center-side positions), and 4 write operations (update the 3 affected row summary values, and the dead/live value of the new cell). This is a slight improvement from 8 reads and 1 write, assuming writes are no slower than reads. I'm guessing you might be able to be more clever with such configurations and arrive at an even better improvement along these lines.
If you don't want anything too complex, then you can use a grid to slice it up, and if that part of the grid is empty, don't try to simulate it (please view Tyler's answer). However, you could do a few optimizations:
Set different grid sizes depending on the amount of live cells, so if there's not a lot of live cells, that likely means they are in a tiny place.
When you randomize it, don't use the grid code until the user changes the data: I've personally tested randomizing it, and even after a long amount of time, it still fills most of the board (unless for a sufficiently small grid, at which point it won't help that much anymore)
If you are showing it to the screen, don't use rectangles for pixel size 1 and 2: instead set the pixels of the output. Any higher pixel size and I find it's okay to use the native rectangle-filling code. Also, preset the background so you don't have to fill the rectangles for the dead cells (not live, because live cells disappear pretty quickly)
Don't exactly know how this can be done, but I remember some of my friends had to represent this game's grid with a Quadtree for a assignment. I'm guess it's real good for optimizing the space of the grid since you basically only represent the occupied cells. I don't know about execution speed though.
It's a two dimensional automaton, so you can probably look up optimization techniques. Your notion seems to be about compressing the number of cells you need to check at each step. Since you only ever need to check cells that are occupied or adjacent to an occupied cell, perhaps you could keep a buffer of all such cells, updating it at each step as you process each cell.
If your field is initially empty, this will be much faster. You probably can find some balance point at which maintaining the buffer is more costly than processing all the cells.
There are table-driven solutions for this that resolve multiple cells in each table lookup. A google query should give you some examples.
I implemented this in C#:
All cells have a location, a neighbor count, a state, and access to the rule.
Put all the live cells in array B in array A.
Have all the cells in array A add 1 to the neighbor count of their
neighbors.
Have all the cells in array A put themselves and their neighbors in array B.
All the cells in Array B Update according to the rule and their state.
All the cells in Array B set their neighbors to 0.
Pros:
Ignores cells that don't need to be updated
Cons:
4 arrays: a 2d array for the grid, an array for the live cells, and an array
for the active cells.
Can't process rule B0.
Processes cells one by one.
Cells aren't just booleans
Possible improvements:
Cells also have an "Updated" value, they are updated only if they haven't
updated in the current tick, removing the need of array B as mentioned above
Instead of array B being the ones with live neighbors, array B could be the
cells without, and those check for rule B0.