On Tue, Nov 18, 2008 at 2:30 AM, Thomas Davie <span dir="ltr"><<a href="mailto:tom.davie@gmail.com">tom.davie@gmail.com</a>></span> wrote:<br><div class="gmail_quote"><div> <snip me><br></div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Wow, that's rather nice, unfortunately I can't run it at the moment, because I've run into the GLUT problems so many people have. I'm wondering though about your definition of randomBehavior.<br>
<br>
randomBehavior :: (Random a) => Double -> Behavior a<br>
randomBehavior s = fmap (fst . random . mkStdGen . round . (+s)) time<br>
<br>
I don't know the theory behind pseudo random number generators well enough to be sure, but I have a feeling that while this may be good enough for a game, it's probably not good enough for anything the relies on the numbers it generates being totally unpredictable. The reason I say that is that as far as I understand it, the guarentee we're given with a pseudo random number generator is that given an output number, the next output number is impossible to predict. I don't think we're given any guarantee that given a monotonically increasing seed, the output of the generator will look particularly different, or be unpredictable.<br>
<br>
Unfortunately, I don't think that I can come up with a better way to define the behavior though. It would be possible to define an Event at a certain interval that splits the random seed at each occurrence, but I can't do better than that.</blockquote>
<div><br>You're absolutely right about randomBehavior and, unfortunately, I was a bit at a loss for what to do to thread through a random number generator.<br><br>I think the basic conclusion I've come to is that it might actually be 'wrong' to want a Behavior of random values if they are to be properly generated from a pseudo-random generator, as the semantics would require you to somehow be using the RandomGen an infinite number of times to get the right see at any instance. Behaviors are supposed to be continuous afterall.<br>
<br>What I think we really want is a way of saying "at every occurrence of an event, we want an 'a'" such that the distribution of a's is pseudorandom. To me, this would mean trying to make Event, already a Monad, an instance of MonadRandom from Cale Gibbard's library of the same name. This doesn't seem like it should be hard, but I'm not clear on it yet.<br>
</div></div><br>I guess it would drop out pretty readily if one had a way to substitute the values of an infinite list sequentially into an Event stream, since then one could take in a RandomGen and use the randoms :: g -> [a] function to create the list & then pair it with the Event. I don't know if there's a way to do that without breaking the Event abstraction, but conceptually it makes sense to me: it's just establishing an isomorphism between infinite lists.<br>
<br>Any other thoughts anyone?<br><br>Cheers,<br>Creighton<br>