[R] Better way to find distances between points in a set?

[Charles C. Berry]

On Wed, 10 Dec 2008, Stavros Macrakis wrote:

> On Tue, Dec 9, 2008 at 11:52 PM, Charles C. Berry <cberry at tajo.ucsd.edu>wrote:
>
>> The 'better way' to do almost anything starts with a reading of the
>> _posting guide_, which reminds you to
>>        Do your homework before posting [Reasons whyfor deleted]]...
>>
>
>
>> Oh yes, if you are too lazy to look up the posting guide URL, the function
>> help.request() will open it for you when you admit that you haven't yet read
>> it (or lead you thru the further steps to prepare a question to this list if
>> you say that you have read it).
>
>
> Charles, I am distressed by the nasty tone of your note.


Well, I am sorry that my reply caused you distress.

I try to provide useful answers. And I often do the 'homework' in the 
posting guide before responding to be sure I provide useful answers.

Sometimes my 'reward' for a post is to receive off-list requests for 
further help that the 'asker' could have easily worked out by following 
the 'Do your homework' steps. And I sometimes get those requests even when 
I point to the excellent self-help resources in my original post.

I welcome advice on how to do a better job and do so in a gentler, kinder 
manner, but I'd like to maintain a focus on strongly encouraging the use 
of help.request(), the posting guide, and all of the other resources that 
are already available.


> Though I certainly
> agree that poor queries on this list are an annoyance, and that the posting
> guide gives useful advice on what to do before posting, I don't think that
> poor queries are a reason to presume bad faith for any but the most flagrant
> correspondents ("here is my problem set, solve it for me"). And though I
> have not been on this list very long, I have been answering questions on
> other lists like this for a long time.
>
> What's more, it seems to me that Witthoft has a legitimate problem.


To which I provided a solution and to which Carl responded asking for 
further clarification.

I responded to that post too - and I'll admit that it could reasonably be 
seen as having a 'snarky' tone - but Carl will know when he reads it what 
he asked to know.


I tried
> to solve it myself, and here is what I come up with:
>
> test <- matrix(rnorm(20),ncol=2)
>
> which.max(dist(x),arr.ind=TRUE)
>     => Error, though which supports arr.ind, which.max doesn't
>        Yes, that is documented, but it violates reasonable expectations.
>
> which(dist(x)==max(dist(x)),arr.ind=TRUE)
>     => 21  oops, that is a vector index; what about arr.ind?
>        Apparently the dist class (return class of dist fnc.)
>        doesn't act as an array here (and which/arr.ind is
>        happy to treat a vector as an array without warning).
>        Is this a problem in dist, or a problem in which?
>        So how do I map back from 21 to 3,7?  Nothing obvious.
>        ? `[`, ? array, and ? dist don't seem to help.
>
> So let's convert to a matrix explicitly:
>
> which(as.matrix(dist(x)==max(dist(x))),arr.ind=TRUE)
>     =>      row col
>         [1,] 21   1
>        Hmm, that's not right; apparently the == flattened the
>        dist class into its base vector, which which/arr.ind
>        helpfully (?) treated as a matrix.
>
> which(as.matrix(dist(x))==max(dist(x)),arr.ind=TRUE)
>    =>        row col
>          7     7   3
>          3     3   7
>        Finally getting useful results; but converting to the array
>        added in the upper triangular part... oh, well, we need to
>        choose *which* maximum anyway.


OK, so I stopped with the response you just gave and didn't take that next 
step to tell Carl to use upper.tri():

 	which( upper.tri( dist(x) ) &
 		as.matrix(dist(x))==max(dist(x)),arr.ind=TRUE)

but help.search("upper") gets it easily enough. So does 
help.search("triangular").


>
> So, as Witthoft said: what's a better, or cleaner way to do all this?


So I gave him my version of a better, cleaner way. If there is a still 
better and even cleaner way, I'd be delighted to see it.

Carl did not ask any of the questions you listed below. And I didn't see 
anything in his post that implied he wanted answers to them.

Chuck

> And
> why aren't R's operators more regular in their behavior?  Why doesn't
> arr.ind work for which.max?  Why doesn't dist(x)==1 return an object of the
> same shape as the original dist (as matrix(...)==1 would have), but instead
> flatten it?  Where does one find the function that maps from a list of array
> indices to the corresponding vector index and vice versa?
>
>             -s
>

Charles C. Berry                            (858) 534-2098
                                             Dept of Family/Preventive Medicine
E mailto:cberry at tajo.ucsd.edu	            UC San Diego
http://famprevmed.ucsd.edu/faculty/cberry/  La Jolla, San Diego 92093-0901