On Tue, Jul 26, 2011 at 3:27 PM, dokondr
<dokondr@gmail.com> wrote:
Thanks for the detailed reply and example!
Using IntMap as a vector seems to be a good idea.
In your example:
1) I would use:
dot = dot' * dot'
dot' = sum . elems . intersectionWith (*)
norm = sum . fmap (**2) . elems
instead of: dot = sum . elems . intersectionWith (*)
norm = (**0.5) . sum . fmap (**2) . elems
Your dot' is a function, so (dot' * dot') wouldn't type check.
2) I don't understand the syntax:cosineSimilarity <$> lookup x space
<*> lookup y space
What are <$> and <*>?
(lookup x space) has type (Maybe something). So does (lookup y space). We are using "applicative functors" to pull out values from (lookup x space) and (lookup y space), apply cosineSimilarity to the values we pulled out, and wrapping it all back up in a Maybe, depending on whether the lookups found something or not.
<$> is exactly the same thing as fmap.
<*> is harder to explain. But it is just plumbing.