Re: [Haskell-cafe] Is anyone working on a sparse matrix library in Haskell?

On 12/1/12 11:58 PM, Kim-Ee Yeoh wrote:

On Sun, Dec 2, 2012 at 10:52 AM, wren ng thornton wrote:

...

My goal for all this is in setting up the type system, not performance. I figure there are other folks who know and care a lot more about the numerical tricks of giving the actual implementations.

You've got my support -- old-school optimizations, numerical, compiler, or otherwise, are deadly boring. Death to them, I say! If we don't explore uncharted waters, who will?

Well, there are interesting things to optimization[1], it's just that that's not my main area and I have few enough round tuits as it is. I also don't spend much time thinking about hardware, but I'm terribly glad there're other folks who really care about it. [1] For example, while matrix multiplication is associative, how exactly you associate things has a major impact on performance. Performance-minded compilers for linear algebra thus choose how to associate things by running an algorithm which is essentially the same as the chart-parsing algorithms in NLP. As a NLPer, I think that's awesome; and since I'm not sure if anyone else has made that connection before, it'd be nice to see what each side could learn from the other. One thing I'd like to get out of the type classes I'm working on is the ability to define a DSL which allows this sort of optimization. -- Live well, ~wren