Get coordinates of centroids

[delanoe]

Hello,
is it possible to get the coordinates only (and not the clusters for instances) ?
Cheers,

[alpmestan]

Hello, (ou bien bonjour =P)

Centroids are not returned by anything exposed to the user at the moment. I would take a patch that does it though if that's useful and if you feel like putting one together? My newborn eagerly consumes all my free time these days =)

[delanoe]

Hello, (oui bonjour pourquoi pas ;)
I was wondering if the coordinates were computed already (then I just need to find the way to give it back) or coordinates are not computed (a quick look on the code let me think it is possible) so I have to calculate it.

Cheers,
(Congratulations for your newborn.)

[alpmestan]

They are computed by centroidsOf, which is called in kmeansStep: https://github.com/alpmestan/kmeans-vector/blob/master/Math/KMeans.hs#L154

You basically want to write a variant of kmeansStep and kmeansWith that doesn't "forget" about the centroids but return them along with the clusters.

[delanoe]

Hello,
I try to add type signature to anonymous functions inside kmeansWith in order to write the variant. I have an idea how I could do it but I encounter some type issues.

So, for instance, if I add type signature to the go function I get the error below (and If i change the type into the expected type I got the same error almost). Do you know why ? Is that the reason why you commented it in the code ? Thanks for your help.

Couldn't match type ‘a1’ with ‘a’
‘a1’ is a rigid type variable bound by
the type signature for go :: Clusters a1 -> Clusters a1
at /kmeans-vector/Math/KMeans.hs:147:11
‘a’ is a rigid type variable bound by
the type signature for
kmeansWith :: Monad m =>
(Int -> [a] -> m (Clusters a))
-> (a -> V.Vector Double)
-> Distance
-> Int
-> [a]
-> m (Clusters a)
at /kmeans-vector/Math/KMeans.hs:137:15
Expected type: G.Vector (Cluster a)
Actual type: Clusters a1
Relevant bindings include
pgroups :: Clusters a1
(bound at /kmeans-vector/Math/KMeans.hs:148:8)
go :: Clusters a1 -> Clusters a1
(bound at /kmeans-vector/Math/KMeans.hs:148:5)
kmeansStep :: G.Vector (Cluster a) -> G.Vector (Cluster a)
(bound at /kmeans-vector/Math/KMeans.hs:154:5)
centroidsOf :: G.Vector (Cluster a) -> G.Vector (V.Vector Double)
(bound at /kmeans-vector/Math/KMeans.hs:163:5)
pairToClosestCentroid :: G.Vector (V.Vector Double)
-> a -> (Int, a)
(bound at /kmeans-vector/Math/KMeans.hs:174:5)
pgroupsEqualUnder :: G.Vector (Cluster a)
-> G.Vector (Cluster a) -> Bool
(bound at /kmeans-vector/Math/KMeans.hs:178:5)
initF :: Int -> [a] -> m (Clusters a)
(bound at /kmeans-vector/Math/KMeans.hs:144:12)
(Some bindings suppressed; use -fmax-relevant-binds=N or -fno-max-relevant-binds)
In the first argument of ‘kmeansStep’, namely ‘pgroups’
In the expression: kmeansStep pgroups
Failed, modules loaded: none.

[alpmestan]

I think you just need to turn the ScopedTypeVariable language extension on, at the top of the module?

{-# LANGUAGE ScopedTypeVariables #-}
module [...] where

I think I left the type sigs commented out just in order to avoid having to turn that extension on (although it's harmless), but I don't remember, it's been a while =)

[delanoe]

Here are the centroids kept in a tuple:

https://github.com/delanoe/kmeans-vector/blob/anoe/master/Math/KMeans.hs#L201

Maybe the tuple is not the right data structure.

I am thinking about something like a data structure keeping both Centroid and Clusters.

Do you have any advice ?

[alpmestan]

Yeah I guess we could make Cluster a data type with an additional field that points to the centroid of the said cluster.