[Rd] matrix bands

[Jeremy David Silver]

Thanks for the suggestion, Martin!

I looked at both Matrix::band and Matrix::bandSparse. Maybe I 
misunderstand the help pages and the examples, but from what I can see 
neither of them provides the functionality I was looking for.

For the getter version of the function I was looking for, I can't use 
Matrix::band (by extracting the non-zero elements) because this would 
require the assumption that the specified band is entirely non-zero, e.g.:

band3 <- function(x,n){
   x <- Matrix::band(x,n,n)
   x[x!=0] ## works for a dense matrix, but not if band n has zeroes
}

As for the setter version of the function I was looking for, it should 
set the values of a specified band in an existing matrix to be the 
vector of values provided. As far as I understand, Matrix::bandSparse 
constructs a matrix rather than modifying the values of an existing matrix.

The functions included in my last post filled this gap. If Matrix::band 
and Matrix::bandSparse can set/get bands in this way, I would like to 
see how this can be done. If not, then perhaps they could be extended to 
achieve this functionality.


On 2011-08-26 14:08, Martin Maechler wrote:
>>>>>> Jeremy David Silver<jds at dmu.dk>
>>>>>>      on Fri, 26 Aug 2011 13:23:43 +0200 writes:
>      >  Dear R developers, I was looking for a function analogous
>      >  to base::diag() for getting and setting bands of a
>      >  matrix. The closest I could find was Matrix::band(), but
>      >  this was not exactly what I wanted for two
>      >  reasons. Firstly, Matrix::band() returns a matrix rather
>      >  than just the specified band.  Secondly, Matrix::band()
>      >  cannot be used for setting the values for a matrix band.
>
> Yes, but did you not look at  help(band)
> or not look carefully enough ?
>
> --->
>
>    See Also:
>
>         ‘bandSparse’ for the _construction_ of a banded sparse matrix
>         directly from its non-zero diagonals.
>
>
>
>      >  Setting or getting a matrix band could be of interest for
>      >  sufficiently many users that you might consider including
>      >  it in base R.  Alternatively, something like this could be
>      >  included in the Matrix package.
>
> well, see above and let us know if you see anything lacking in
> bandSparse().
> Till now we haven't got much feedback about it, and there may
> well be room for improvement.
>
> Martin Maechler, ETH Zurich  and co- maintainer("Matrix")
>