I seem to be having confusion at the runRVar level of random-fu. I can't figure out how to use the Data.Random.Source.PureMT module to get a meaningful random source (I can't get my code to type-check). I wrote a [trivial] flipCoin function
flipCoin = uniform False True and am trying to fill in the final place of runRVar :t runRVar (replicateM 20 flipCoin) runRVar (replicateM 20 flipCoin) :: (RandomSource m s) => s -> m [Bool]
-- Alex R
A PureMT generator is immutable, so must be threaded through the monad in which you are sampling. There are RandomSource instances provided for a few special cases, including "IORef PureMT" in the IO monad. For example: main = do mt <- newPureMT src <- newIORef mt flips <- runRVar (replicateM 20 flipCoin) src print flips Alternatively, the functions in the module you mentioned can be used to define additional instances, such as: instance MonadRandom (State PureMT) where supportedPrims _ _ = True getSupportedRandomPrim = getRandomPrimFromPureMTState And RandomSource instances look almost the same. See the Data.Random.Source.PureMT source for examples. (I thought I had included this particular instance in the distribution but I apparently missed it. The next release will probably include this as well as corresponding instances for the 'transformers' package, possibly separated out into 'random-fu-mtl' and 'random-fu-transformers' packages). The "StdRandom" type is a convenient "RandomSource" designating this instance in the State PureMT monad. Personally, I prefer to use the "sample" function for this purpose, as well as the "sampleFrom" function in place of runRVar/runRVarT. GHCi does not display the "sample" functions' types properly - they are defined for RVarT as well as for all Distribution instances. Sorry it took so long responding. -- James On Sep 2, 2010, at 10:01 AM, Alex Rozenshteyn wrote:
I seem to be having confusion at the runRVar level of random-fu.
I can't figure out how to use the Data.Random.Source.PureMT module to get a meaningful random source (I can't get my code to type-check).
I wrote a [trivial] flipCoin function
flipCoin = uniform False True and am trying to fill in the final place of runRVar :t runRVar (replicateM 20 flipCoin) runRVar (replicateM 20 flipCoin) :: (RandomSource m s) => s -> m [Bool]
-- Alex R
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Okay, I figured the immutability bit out and I got the IORef example working, but I can't get it to work with state.
put (pureMT 0) >>= runRVar flipCoin
gives me two type errors: "No instance for (MonadState PureMT m)" and "No instance for (RandomSource m ())"
runState $ put (pureMT 0) >>= runRVar flipCoin runState $ put (pureMT 0) >> get >>= runRVar flipCoin put (pureMT 0) >> get >>= runRVar flipCoin
and other desperate attempts, some of which in hindsight are too
embarrassing to list give me similar errors. I'm trying to do figure out
how to do this without going to the IO monad (so I can run it with the same
seed to replicate results).
On Tue, Sep 7, 2010 at 3:14 PM, James Andrew Cook
A PureMT generator is immutable, so must be threaded through the monad in which you are sampling. There are RandomSource instances provided for a few special cases, including "IORef PureMT" in the IO monad. For example:
main = do mt <- newPureMT src <- newIORef mt flips <- runRVar (replicateM 20 flipCoin) src print flips
Alternatively, the functions in the module you mentioned can be used to define additional instances, such as:
instance MonadRandom (State PureMT) where supportedPrims _ _ = True getSupportedRandomPrim = getRandomPrimFromPureMTState
And RandomSource instances look almost the same. See the Data.Random.Source.PureMT source for examples. (I thought I had included this particular instance in the distribution but I apparently missed it. The next release will probably include this as well as corresponding instances for the 'transformers' package, possibly separated out into 'random-fu-mtl' and 'random-fu-transformers' packages).
The "StdRandom" type is a convenient "RandomSource" designating this instance in the State PureMT monad. Personally, I prefer to use the "sample" function for this purpose, as well as the "sampleFrom" function in place of runRVar/runRVarT. GHCi does not display the "sample" functions' types properly - they are defined for RVarT as well as for all Distribution instances.
Sorry it took so long responding.
-- James
On Sep 2, 2010, at 10:01 AM, Alex Rozenshteyn wrote:
I seem to be having confusion at the runRVar level of random-fu.
I can't figure out how to use the Data.Random.Source.PureMT module to get a meaningful random source (I can't get my code to type-check).
I wrote a [trivial] flipCoin function
flipCoin = uniform False True and am trying to fill in the final place of runRVar :t runRVar (replicateM 20 flipCoin) runRVar (replicateM 20 flipCoin) :: (RandomSource m s) => s -> m [Bool]
-- Alex R
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
-- Alex R
On Sep 7, 2010, at 10:21 AM, Alex Rozenshteyn wrote:
Okay, I figured the immutability bit out and I got the IORef example working, but I can't get it to work with state.
put (pureMT 0) >>= runRVar flipCoin
gives me two type errors: "No instance for (MonadState PureMT m)" and "No instance for (RandomSource m ())"
The first error is because "put" has the general type: "put :: (MonadState s m) => s -> m ()" and GHC doesn't know what monad you want to evaluate it in. Just off the bat there are 2 possibilities provided by Control.Monad.State: State and StateT. The second is because runRVar requires an additional argument specifying the source from which to sample. In your case, you want to sample from the "standard" source, the MonadRandom instance. The type "StdRandom" with a single constructor of the same name designates the MonadRandom instance. So for your example, a working incantation would be:
put (pureMT 0) >>= runRVar flipCoin StdRandom :: State PureMT Bool
To actually run this action you would then use runState or evalState, in which case the type annotation would no longer be necessary because the use of runState or evalState would give the compiler enough information to know what you want. These functions both also accept an initial state as an argument, so you don't actually need "put" either:
evalState (runRVar flipCoin StdRandom) (pureMT 0)
Using the "sample" function I mentioned earlier you can leave off the mention of StdRandom:
evalState (sample flipCoin) (pureMT 0)
I'm trying to do figure out how to do this without going to the IO monad (so I can run it with the same seed to replicate results).
Incidentally, you can use the 'pureMT' function to seed your generator in the IO monad just as easily:
newIORef (pureMT 0) >>= \src -> runRVar flipCoin src
or
do src <- newIORef (pureMT 0); runRVar flipCoin src
A large part of the point of the RVar monad as its own independent construct is to allow you to use random variables such as flipCoin in any monad that can support them while at the same time guaranteeing all the same purity / safety as if you had used something "obviously" pure such as State. -- James
participants (2)
-
Alex Rozenshteyn -
James Andrew Cook