[Haskell-cafe] statistics package and randomness
I'm trying to learn how to use randomness in Haskell and it seems very non-straightforward and complex. I could do a lot of things using 'split' from System.Random, but apparently it's broken. There is the statistics package here: http://hackage.haskell.org/package/statistics Is this a better solution? It uses the ST monad in the RandomVariate module. Can someone point me to a tutorial explaining ST, and/or a tutorial in the RandomVariate module? Pseudorandomness seems like one case where it would just be a hell of a lot simpler to have a global generator--never split the state. Is the ST monad some way to accomplish this? Thanks, Mike ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] statistics package and randomness
i'll try a very non-technical explanation that has worked for me so far. (is it correct? does it make sense?) IO and ST are quite similar. the difference is that whereas IO gives you a concept of time in the world surrounding your code, ST lets you create a little bubble inside your code in which you can maintain state, while the bubble as a whole acts all pure and lazy. for example, if you want to implement an algorithm that writes to and reads from a matrix, you use ST: you want to control the order in which you read from and write to it, but not the order in which access events to that data structure mixes with user interaction events. -matthias On Mon, Oct 12, 2009 at 12:25:43AM -0700, Michael Mossey wrote: To: Haskell Cafe Haskell-Cafe@haskell.org Cc: From: Michael Mossey m...@alumni.caltech.edu Date: Mon, 12 Oct 2009 00:25:43 -0700 Subject: [Haskell-cafe] statistics package and randomness I'm trying to learn how to use randomness in Haskell and it seems very non-straightforward and complex. I could do a lot of things using 'split' from System.Random, but apparently it's broken. There is the statistics package here: http://hackage.haskell.org/package/statistics Is this a better solution? It uses the ST monad in the RandomVariate module. Can someone point me to a tutorial explaining ST, and/or a tutorial in the RandomVariate module? Pseudorandomness seems like one case where it would just be a hell of a lot simpler to have a global generator--never split the state. Is the ST monad some way to accomplish this? Thanks, Mike ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe ** ACCEPT: CRM114 PASS osb unique microgroom Matcher ** CLASSIFY succeeds; success probability: 1. pR: 5.5394 Best match to file #0 (nonspam.css) prob: 1. pR: 5.5394 Total features in input file: 2960 #0 (nonspam.css): features: 758386, hits: 2888631, prob: 1.00e+00, pR: 5.54 #1 (spam.css): features: 1683715, hits: 3150692, prob: 2.89e-06, pR: -5.54 ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] statistics package and randomness
On Mon, Oct 12, 2009 at 12:25 AM, Michael Mossey m...@alumni.caltech.eduwrote: I'm trying to learn how to use randomness in Haskell and it seems very non-straightforward and complex. I could do a lot of things using 'split' from System.Random, but apparently it's broken. There is the statistics package here: http://hackage.haskell.org/package/statistics Is this a better solution? Yes, as it's much faster, more robust, and (depending on your perspective) easier to use, provided you understand the ST monad. It uses the ST monad in the RandomVariate module. Can someone point me to a tutorial explaining ST, and/or a tutorial in the RandomVariate module? For a tutorial on grokking ST, I'd suggest chapter 26 of Real World Haskell, but I'm biased, since I wrote it: http://book.realworldhaskell.org/read/advanced-library-design-building-a-bloom-filter.html Note that this already assumes that you understand monads pretty well. Pseudorandomness seems like one case where it would just be a hell of a lot simpler to have a global generator--never split the state. Is the ST monad some way to accomplish this? Having ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: [Haskell-cafe] statistics package and randomness
On Mon, Oct 12, 2009 at 11:01 AM, Bryan O'Sullivan b...@serpentine.comwrote: Pseudorandomness seems like one case where it would just be a hell of a lot simpler to have a global generator--never split the state. Is the ST monad some way to accomplish this? Having [...] Feh, gmail fail. Having a global generator is not actually a good thing, since it has to live *somewhere*. If you keep its existence implicit, it becomes slow, since you have to lock it against concurrent use by multiple threads. If you make it explicit, you have to plumb the thing all over the place by hand, which is also nasty. The advantage of putting the PRNG in the ST monad is that you can seed a new PRNG close to the point where you'll need it, and not need to pass around so much state. ___ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
