On Tue, Jan 13, 2009 at 3:29 PM, Phil <span dir="ltr"><<a href="mailto:pbeadling@mail2web.com">pbeadling@mail2web.com</a>></span> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
My only concern with using this method is - Will 'iterate' not create a full<br>
list of type [Double] and then take the final position once the list has<br>
been fully realized? For my application this would be undesirable as the<br>
list may be millions of items long, and you only ever care about the last<br>
iteration (It's a crude Monte Carlo simulator to give it some context). If<br>
Haskell is smart enough to look ahead and see as we only need the last<br>
element as it is creating the list, therefore garbage collecting earlier<br>
items then this would work fine - by I'm guessing that is a step to far for<br>
the compiler?</blockquote><div><br>No, doing this type of thing is very typical Haskell, and the garbage collector <i>will</i> incrementally throw away early elements of the list. <br><br></div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
<br>
I had originally implemented this similar to the above (although I didn't<br>
know about the 'iterate' keyword</blockquote><div><br>FWIW, iterate is just a function, not a keyword. Could just be terminology mismatch.<br> <br>So, while the garbage collector will do the right thing, for a list millions of elements long, I suspect you will get stack overflows and/or bad memory performance because the computation is too lazy. One solution is to use a stricter version of !!, which evaluates elements of the list as it whizzes by them. Because the function you're iterating is strict to begin with, you do not lose performance by doing this:<br>
<br><font face="courier new,monospace">strictIdx :: Int -> [a] -> a<br>strictIdx _ [] = error "empty list"<br>strictIdx 0 (x:xs) = x<br>strictIdx n (x:xs) = x `seq` strictIdx (n-1) xs<br></font><br>(Note that I flipped the arguments, to an order that is nicer for currying)<br>
<br>The reason is that iterate f x0 constructs a list like this:<br><br>[ x0, f x0, f (f x0), f (f (f x0)), ... ]<br><br>But shares the intermediate elements, so if we were to evaluate the first f x0 to, say, 42, then the thunks are overwritten and become:<br>
<br>[ x0, 42, f 42, f (f 42), ... ]<br><br>So iterate f x0 !! 1000000 is f (f (f (f ( ... a million times ... f x0)))), which will be a stack overflow because of each of the calls. What strictIdx does is to evaluate each element as it traverses it, so that each call is only one function deep, then we move on to the next one.<br>
<br>This is the laziness abstraction leaking. Intuition about it develops with time and experience. It would be great if this leak could be patched by some brilliant theorist somewhere.<br><br>Luke<br><br></div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
- which makes things tidier - a useful<br>
tip!), I moved to using the state monad and replicateM_ for the first<br>
truncate(endTime/timeStep)-1 elements so that everything but the last result<br>
is thrown away, and a final bind to getEvolution would return the result.<br>
<br>
Now that the code has been modified so that no result is passed back, using<br>
modify and execState, this can be simplified to "replicateM_<br>
truncate(endTime/timeStep)" with no final bind needed. I've tried this and<br>
it works fine.<br>
<br>
The key reason for using the Monad was to tell Haskell to discard all but<br>
the current state. If I'm wrong about please let me know, as I don't want<br>
to be guilty of overcomplicating my algorithm, and more importantly it means<br>
I'm not yet totally grasping the power of Haskell!<br>
<br>
Thanks again,<br>
<font color="#888888"><br>
Phil.<br>
</font><div><div></div><div class="Wj3C7c"><br>
<br>
<br>
<br>
On 13/01/2009 03:13, "David Menendez" <<a href="mailto:dave@zednenem.com">dave@zednenem.com</a>> wrote:<br>
<br>
> On Mon, Jan 12, 2009 at 8:34 PM, Phil <<a href="mailto:pbeadling@mail2web.com">pbeadling@mail2web.com</a>> wrote:<br>
>> Thanks Minh - I've updated my code as you suggested. This looks better than<br>
>> my first attempt!<br>
>><br>
>> Is it possible to clean this up any more? I find:<br>
>><br>
>> ( (), (Double, Word64) )<br>
>><br>
>> a bit odd syntactically, although I understand this is just to fit the type<br>
>> to the State c'tor so that we don't have to write our own Monad longhand.<br>
><br>
> If you have a function which transforms the state, you can lift it<br>
> into the state monad using "modify".<br>
><br>
>> evolveUnderlying :: (Double, Word64) -> (Double, Word64)<br>
>> evolveUnderlying (stock, state) = ( newStock, newState )<br>
>> where<br>
>> newState = ranq1Increment state<br>
>> newStock = stock * exp ( ( ir - (0.5*(vol*vol)) )*timeStep + (<br>
>> vol*sqrt(timeStep)*normalFromRngState(state) ) )<br>
>><br>
>> getEvolution :: State (Double, Word64) ()<br>
>> getEvolution = modify evolveUnderlying<br>
><br>
> Now, I don't know the full context of what you're doing, but the<br>
> example you posted isn't really gaining anything from the state monad.<br>
> Specifically,<br>
><br>
> execState (replicateM_ n (modify f))<br>
> = execState (modify f >> modify f >> ... >> modify f)<br>
> = execState (modify (f . f . ... . f))<br>
> = f . f . ... . f<br>
><br>
> So you could just write something along these lines,<br>
><br>
>> mcSimulate :: Double -> Double -> Word64 -> [Double]<br>
>> mcSimulate startStock endTime seedForSeed = fst expiryStock : mcSimulate<br>
>> startStock endTime newSeedForSeed<br>
>> where<br>
>> expiryStock = iterate evolveUnderlying (startStock, ranq1Init seedForSeed)<br>
>> !! truncate (endTime/timeStep)<br>
>> newSeedForSeed = seedForSeed + 246524<br>
><br>
><br>
> Coming back to your original question, it is possible to work with<br>
> nested state monad transformers. The trick is to use "lift" to make<br>
> sure you are working with the appropriate state.<br>
><br>
> get :: StateT s1 (State s2) s1<br>
> put :: s1 -> StateT s1 (State s2) ()<br>
><br>
> lift get :: StateT s1 (State s2) s2<br>
> lift put :: s2 -> StateT s1 (State s2) ()<br>
><br>
> A more general piece of advice is to try breaking things into smaller<br>
> pieces. For example:<br>
><br>
> getRanq1 :: MonadState Word64 m => m Word64<br>
> getRanq1 = do<br>
> seed <- get<br>
> put (ranq1Increment seed)<br>
> return seed<br>
><br>
> getEvolution :: StateT Double (State Word64) ()<br>
> getEvolution = do<br>
> seed <- lift getRanq1<br>
> modify $ \stock -> stock * exp ( ( ir - (0.5*(vol*vol)) )*timeStep<br>
> + ( vol*sqrt(timeStep)*normalFromRngState(seed) ) )<br>
><br>
<br>
</div></div><div><div></div><div class="Wj3C7c">_______________________________________________<br>
Haskell-Cafe mailing list<br>
<a href="mailto:Haskell-Cafe@haskell.org">Haskell-Cafe@haskell.org</a><br>
<a href="http://www.haskell.org/mailman/listinfo/haskell-cafe" target="_blank">http://www.haskell.org/mailman/listinfo/haskell-cafe</a><br>
</div></div></blockquote></div><br>