Re: [Haskell-cafe] Array Binary IO & molecular simulation
To sum up here is the example that can write two arrays in one file and then read this two arrays back. To restore written data it just reads the file into bytestring, then splits the bytestring into equal parts. The parts are decoded. I suppose the method is suitable for decoding files with unboxed arrays of equal size. import Data.Array.Unboxed import Data.Binary import qualified Data.ByteString.Lazy as BL import IO a = listArray ((1,1),(3,2)) [3,4,5,6,7,8] :: UArray (Int, Int) Float b = listArray ((1,1),(3,2)) [9,10,11,12,13,14] :: UArray (Int, Int) Float encodeFile2 f = BL.appendFile f . encode encoder = do encodeFile "Results.txt" a encodeFile2 "Results.txt" b decoder = do contents <- BL.readFile "Results.txt" print $ (show (decode (fst (BL.splitAt 118 contents)) :: UArray (Int, Int) Float)) print $ (show (decode (snd (BL.splitAt 118 contents)) :: UArray (Int, Int) Float)) P.S. I've already sent this letter to mailist several ours ago, but it wasn't published :-/ ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] Array Binary IO & molecular simulation
To sum up here is the example that can write two arrays in one file and then read this two arrays back. To restore written data it just reads the file into bytestring, then splits the bytestring into equal parts. The parts are decoded. I suppose the method is suitable for decoding files with unboxed arrays of equal size. import Data.Array.Unboxed import Data.Binary import qualified Data.ByteString.Lazy as BL import IO a = listArray ((1,1),(3,2)) [3,4,5,6,7,8] :: UArray (Int, Int) Float b = listArray ((1,1),(3,2)) [9,10,11,12,13,14] :: UArray (Int, Int) Float encodeFile2 f = BL.appendFile f . encode encoder = do encodeFile "Results.txt" a encodeFile2 "Results.txt" b decoder = do contents <- BL.readFile "Results.txt" print $ (show (decode (fst (BL.splitAt 118 contents)) :: UArray (Int, Int) Float)) print $ (show (decode (snd (BL.splitAt 118 contents)) :: UArray (Int, Int) Float)) ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] Array Binary IO & molecular simulation
2009/05/02 Grigory Sarnitskiy : > 2009/05/02 Jason Dusek : >> The original poster should try serializing a tuple of arrays >> instead of serializing each array individually. > > Maybe, but I have some doubts. I have to operate with > thousands of arrays --- are tuples good in such case? No, they can be only 17 elements long. Then you need lists of arrays, I guess. > Moreover it is desirable to write data as it is calculated to > be sure it wont be lost... Would it be too much trouble to write to separate files? -- Jason Dusek ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] Array Binary IO & molecular simulation
> 2009/05/02 Jason Dusek :>The original poster should try serializing a tuple of> arrays instead of serializing each array individually.Maybe, but I have some doubts. I have to operate with thousands of arrays --- are tuples good in such case? Moreover it is desirable to write data as it is calculated to be sure it wont be lost if the program suddenly stops for some reason. Can tuple/array/list of arrays be serialized and written to the file without its whole computation, but just as new elements appear? ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] Array Binary IO & molecular simulation
2009/05/02 Krzysztof Skrzętnicki : > 2009/05/02 Jason Dusek : >> I believe Data.Binary includes a header with the data it >> serializes. In consequence, the second and all following >> arrays will be invisible. > > I didn't check the Binary instance for arrays, however I > belive it's not true. Consider Binary instances for tuples: These instances describe serializers for tuples; however, there is nothing preventing Data.Binary from tacking on a length header or anything else like that when `encode` is called. The original poster should try serializing a tuple of arrays instead of serializing each array individually. > If what you are saying would be true, the tuple instances > would be incorrect for array tuples. What is an array tuple? -- Jason Dusek ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] Array Binary IO & molecular simulation
On Sat, May 2, 2009 at 11:12, Jason Dusek wrote: > I believe Data.Binary includes a header with the data > it serializes. In consequence, the second and all following > arrays will be invisible. [Apologies for Jason for sending this twice to him] I didn't check the Binary instance for arrays, however I belive it's not true. Consider Binary instances for tuples: instance (Binary a, Binary b) => Binary (a,b) where put (a,b) = put a >> put b get = liftM2 (,) get get instance (Binary a, Binary b, Binary c) => Binary (a,b,c) where put (a,b,c) = put a >> put b >> put c get = liftM3 (,,) get get get If what you are saying would be true, the tuple instances would be incorrect for array tuples. Best regards Christopher Skrzętnicki ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] Array Binary IO & molecular simulation
I believe Data.Binary includes a header with the data it serializes. In consequence, the second and all following arrays will be invisible. Sorry I can not be of more help. -- Jason Dusek ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] Array Binary IO & molecular simulation
So I wonder of existing projects of such type, both Molecular dynamics and Monte Carlo methods. The fastest Haskell Monte Carlo code I've seen in action is Simon's port of a Monte Carlo Go engine: http://www.haskell.org/pipermail/haskell-cafe/2009-March/057982.html http://www.haskell.org/pipermail/haskell-cafe/2009-March/058069.html That is competitive to lightly optimised non-Haskell versions, though not competitive with highly optimised versions (the kind that slows down when you update your compiler, but gives you dramatic boosts due to detailed attention both to generated assembly code and to high-level algorithm shortcuts). Though there's also specialization as an option http://www.cse.unsw.edu.au/~chak/papers/KCCSB07.html and googling for "haskell monte carlo" give a few more hits, such as http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.47.2981 http://www.citeulike.org/user/dguibert/article/561815 and even some hackage package, though I don't know whether efficiency was a concern there? I've got also some technical questions. Now I'm using 2D DiffUArray to represent particle positions during the simulation (when there are lots of array updates). Is this reasonably fast (I want to use pure external interface of DiffArray)? DiffArray is slow (don't know about DiffUArray): http://hackage.haskell.org/trac/ghc/ticket/2727 The default Random is also slow (see the mersenne alternatives on hackage instead, and be careful if you use them through the standard class interface): http://hackage.haskell.org/trac/ghc/ticket/2280 http://hackage.haskell.org/trac/ghc/ticket/427 Claus ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
