I do prefer my solution so far though. Transposing is a very obvious solution to the problem and if anyone needed to read the code afterwards I think the solution with transpose would be digested the fastest. Yup that’s the maintenance coder coming out in me.

Anyway, thanks for taking the time to do the profiling, and the commenting

@Axman6: Thanks for commenting! I agree, it does look like it’d be nice and quick. If and when I get the chance to profile it I’ll put some stats up. Unless you’ve already profiled it…???

@Dan: That’s a great question One that I’ve spent a few days thinking about because I wasn’t sure exactly what the answer was. To be honest, I’m not sure I do now! But I’ll give it a crack. Pardon me if I don’t articulate it very well.

The first part of the problem is that most people who are writing code that lacks performance don’t know what they don’t know. It’s becoming more and more common for non-developers, or inexperienced developers, to build production level systems in code without really having a full grasp of what’s required to write high quality and efficient software. That wasn’t intended to be a stab That’s a gross generalisation!

So people will go through their formal education in their own areas and then attempt to fill whatever gaps they have in the real world with their untrained efforts.

The perfect parallel to this is programmer art! Everyone knows what happens when a developer attempts to create a user interface. It’s abysmal. The colours are bad, there’s no “flow”, it’s abrasive to use and the UI looks awful. It’s a typical case of a programmer thinking they can design. Most of the time, they can’t. They just think they can! Thankfully, I am in the boat where I know I can’t do design I can write you an interface that works, but don’t expect it to be nice! That’s where I aim to utilise other people’s talents.

So I guess what I’m saying is that without a good background in development or some formal training in algorithms, you’re going to find it hard because the points where there are inefficiencies aren’t going to be immediately obvious to you.

Having said all that, there are definitely things you can do to aid you in making your systems function better. These are the kinds of things that anyone can do. Those things are the content of another blog post I’m inspired to write them up in a more formal way and actually try to do a better job of justifying what I’ve said here.

So without sounding cheezy, watch this space! I’ll have a solid answer for you very shortly.

extremes = (\(y:ys) -> foldl' (\(l,h) x -> (min l x, max h x)) (y,y) ys) . foldl (zipWith (+)) (repeat 0)

Seems I had a lot of the same ideas as Dougal. I think that at least the left foldl’ should be optimised pretty nicely.
– Axman6

e1 = (maximum &&& minimum) . map sum . transpose
e2 = foldl1 (curry (uncurry max . fst &&& uncurry min . snd)) . join zip . map sum . transpose
e3 = foldl1 (curry (uncurry max . fst &&& uncurry min . snd)) . join zip . foldr (zipWith (+)) (repeat 0)

… and they all came up around the 3s mark, jostling for top spot depending on how many times I can them.

I think this might mean that GHC is clever enough to prune the intermediate matrix? Or this could be just an effect of lazy evaluation (effectively doing the pruning automatically). Hard to tell really.

@Dan – The haskell profiling stuff is pretty good, easiest way to find out if something is going wrong (wrong order polynomial) is to run it a bunch of times and graph it!

@jberryman: You raise a good point I am yet to profile any of the solutions mentioned above. That is something I’ll aim to do at some stage today or tomorrow and I’ll post the results as an edit to the blog post. Thanks for the feedback!