Friends

 

Random number generation lies far outside my expertise, but I know it to be an area where it’s easy to mess up, either in performance or in generating genuinely random numbers.  So I’m delighted that Dominic and his colleagues have been working so hard on this.   We all owe them a debt of thanks.  Good RNGs are at the beating heart of many other algorithms, but are rather un-loved as an object of study in their own right.

 

So thank you Dominic, @curiousleo @lehins  -- and indeed Guy Steele and colleagues, on whose work this is based.  We don’t often get perf boosts of 1000x!

 

I hope that that, after suitable scrutiny and refinement if necessary, this ends up being accepted.

 

Simon

 

From: Libraries <libraries-bounces@haskell.org> On Behalf Of dominic@steinitz.org
Sent: 26 May 2020 11:00
To: libraries <libraries@haskell.org>
Cc: Alexey Kuleshevich <alexey@kuleshevi.ch>
Subject: Improving Random

 

Hello Libraries,

You may recall that following the blog post by @lehins, a group of us (@curiousleo, @lehins and me) invited participation in February to take this work and apply it to improving the current random library.

Our proximate goals were to fix #25 (filed in 2015) and #51 (filed in 2018). After a lot of discussion and experimentation, we have a proposal that addresses both these issues and also: #26, #44, #53, #55, #58 and #59.

For backwards compatibility, the proposal retains the old style classes and enhances them. Thus in 1.1 we have

class RandomGen g where

    next :: g -> (Int, g)

    genRange :: g -> (Int, Int)

    split :: g -> (g, g)

    {-# MINIMAL next, split #-}

and in 1.2 we have

class RandomGen g where

    next :: g -> (Int, g)

    genWord8 :: g -> (Word8, g)

    genWord16 :: g -> (Word16, g)

    genWord32 :: g -> (Word32, g)

    genWord64 :: g -> (Word64, g)

    genWord32R :: Word32 -> g -> (Word32, g)

    genWord64R :: Word64 -> g -> (Word64, g)

    genShortByteString :: Int

        -> g -> (Data.ByteString.Short.Internal.ShortByteString, g)

    genRange :: g -> (Int, Int)

    split :: g -> (g, g)

    {-# MINIMAL split, (genWord32 | genWord64 | next, genRange) #-}

and next and genRange are deprecated. This interface is what allows the significantly faster performance as no longer is everything forced to go via Integer.

Several new interfaces are introduced and it is recommended that new applications use these and, where feasible, existing applications migrate to using them.

The major API addition in this PR is the definition of a new class MonadRandom:

-- | 'MonadRandom' is an interface to monadic pseudo-random number generators.

class Monad m => MonadRandom g s m | g m -> s where

{-# MINIMAL freezeGen,thawGen,(uniformWord32|uniformWord64) #-}

    type Frozen g = (f :: Type) | f -> g

    freezeGen :: g s -> m (Frozen g)

    thawGen :: Frozen g -> m (g s)

    uniformWord32 :: g s -> m Word32 -- default implementation in terms of uniformWord64

    uniformWord64 :: g s -> m Word64 -- default implementation in terms of uniformWord32

    -- plus methods for other word sizes and for byte strings

    -- all have default implementations so the MINIMAL pragma holds

Conceptually, in MonadRandom g s m, g s is the type of the generator, s is the state type, and m the underlying monad. Via the functional dependency g m -> s, the state type is determined by the generator and monad.

Frozen is the type of the generator's state "at rest". It is defined as an injective type family via f -> g, so there is no ambiguity as to which g any Frozen g belongs to.

This definition is generic enough to accommodate, for example, the Gen type from mwc-random, which itself abstracts over the underlying primitive monad and state token. This is the full instance declaration (provided here as an example - this instance is not part of random as random does not depend on mwc-random):

instance (s ~ PrimState m, PrimMonad m) => MonadRandom MWC.Gen s m where

    type Frozen MWC.Gen = MWC.Seed

    freezeGen = MWC.save

    thawGen = MWC.restore

    uniformWord8 = MWC.uniform

    uniformWord16 = MWC.uniform

    uniformWord32 = MWC.uniform

    uniformWord64 = MWC.uniform

    uniformShortByteString n g = unsafeSTToPrim (genShortByteStringST n (MWC.uniform g))

Pure random number generators can also be made instances of this class providing a uniform interface to both pure and stateful random number generators. An instance for the standard number generator StdGen is provided.

The Random typeclass has conceptually been split into Uniform and UniformRange. The Random typeclass is still included for backwards compatibility. Uniform is for types where it is possible to sample from the type's entire domain; UniformRange is for types where one can sample from a specified range:

class Uniform a where

    uniformM :: MonadRandom g s m => g s -> m a

class UniformRange a where

    uniformRM :: MonadRandom g s m => (a, a) -> g s -> m a

The proposal is a breaking change but the changes are not very intrusive and we have PRs ready for the affected downstream libraries:

• requires base >= 4.10 (GHC-8.2)
• StdGen is no longer an instance of Read
• randomIO and randomRIO were extracted from the Random class into separate functions

In addition, there may be import clashes with new functions, e.g. uniform and uniformR.

Further explanatory details may be found here and the PR for the proposed new version is here.

Here are some benchmarks run on a 3.1 GHz Intel Core i7. The full benchmarks can be run using e.g. stack bench. The benchmarks are measured in milliseconds per 100,000 generations. In some cases, the performance is over x1000(!) times better; the minimum performance increase for the types listed below is more than x35.

| Name                    | Mean (1.1) | Mean (1.2) | Improvement|

| ----------------------- | ---------- | ---------- | ---------- |

| pure/random/Float       |         30 |       0.03 |        1038|

| pure/random/Double      |         52 |       0.03 |        1672|

| pure/random/Integer     |         43 |       0.33 |         131|

| pure/uniform/Word8      |         14 |       0.03 |         422|

| pure/uniform/Word16     |         13 |       0.03 |         375|

| pure/uniform/Word32     |         21 |       0.03 |         594|

| pure/uniform/Word64     |         42 |       0.03 |        1283|

| pure/uniform/Word       |         44 |       0.03 |        1491|

| pure/uniform/Int8       |         15 |       0.03 |         511|

| pure/uniform/Int16      |         15 |       0.03 |         507|

| pure/uniform/Int32      |         22 |       0.03 |         749|

| pure/uniform/Int64      |         44 |       0.03 |        1405|

| pure/uniform/Int        |         43 |       0.03 |        1512|

| pure/uniform/Char       |         17 |       0.49 |          35|

| pure/uniform/Bool       |         18 |       0.03 |         618|

| pure/uniform/CChar      |         14 |       0.03 |         485|

| pure/uniform/CSChar     |         14 |       0.03 |         455|

| pure/uniform/CUChar     |         13 |       0.03 |         448|

| pure/uniform/CShort     |         14 |       0.03 |         473|

| pure/uniform/CUShort    |         13 |       0.03 |         457|

| pure/uniform/CInt       |         21 |       0.03 |         737|

| pure/uniform/CUInt      |         21 |       0.03 |         742|

| pure/uniform/CLong      |         43 |       0.03 |        1544|

| pure/uniform/CULong     |         42 |       0.03 |        1460|

| pure/uniform/CPtrdiff   |         43 |       0.03 |        1494|

| pure/uniform/CSize      |         43 |       0.03 |        1475|

| pure/uniform/CWchar     |         22 |       0.03 |         785|

| pure/uniform/CSigAtomic |         21 |       0.03 |         749|

| pure/uniform/CLLong     |         43 |       0.03 |        1554|

| pure/uniform/CULLong    |         42 |       0.03 |        1505|

| pure/uniform/CIntPtr    |         43 |       0.03 |        1476|

| pure/uniform/CUIntPtr   |         42 |       0.03 |        1463|

| pure/uniform/CIntMax    |         43 |       0.03 |        1535|

| pure/uniform/CUIntMax   |         42 |       0.03 |        1493|

 

Dominic Steinitz

dominic@steinitz.org

http://idontgetoutmuch.org

Twitter: @idontgetoutmuch

 

 

 

 

_______________________________________________
Libraries mailing list
Libraries@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/libraries