The third problem (not passing the state everywhere) got solved by using lift. Thanks to the people from #haskell channel. Interestingly, the code became almost twice as slow as a result..

The new code can be found here: http://pastebin.com/hkeJECem

On Fri, Aug 7, 2015 at 6:17 PM, Ilya Razenshteyn <ilyaraz@gmail.com> wrote:
quickfix: 10M'th, not 1M'th of course.


Cheers,
Ilya

On Fri, Aug 7, 2015 at 6:11 PM, Ilya Razenshteyn <ilyaraz@gmail.com> wrote:
Hi everyone,

I'm new to Haskell and decided to learn a bit about monads and their utility. Actually, what worked nicely for me was: to read first several pages from "Computational lambda-calculus and monads", then do exercises from https://tonymorris.github.io/blog/posts/20-intermediate-haskell-exercises/, and then map these exercises to actual functions in the standard library.

Then, I decided to implement memoization to practice a bit with ST, HashTable's and such. Here is what I managed to get: http://pastebin.com/79pwjLPL . This code computes Fibonacci numbers while caching the intermediate values in a hash table. I tried to make the code as fast as possible (it runs in under 15 seconds, when computing 1M'th Fibonacci number modulo 999983). It uses the package hashtables.

I have several question about all this.

First, is there a cheap way to speed this code up? Note that I'm interested in a universal memoization strategy, that is I don't want to use arrays instead of hash tables etc.

Second, is this code the right way of doing what it does (assuming that I really care about the performance)?

Third question. Currently, the code passes "cache" everywhere. It would be natural to combine ST with something like ReaderT to store it there. What is a way to do it? I tried something like "ReaderT (HashTable s) (ST s) Int", but I have not managed to get it work.