R help - Aug 2005 - matrix conformity with matrix 1x1 and scalars

Hello,
I am calculating this thing with vectors (b) and matrices (G,P):
b'G/sqrt(b'Pb)
where the denominator is a quadratic form and therefore always a scalar.

In Scilab, it is quite simple:
b'*G/sqrt(b'*P*b)
However, in R, the denominator is an (1x1)matrix and R claims it is non
conformable and I have to use drop() or as.numeric(). Like
this:
> b = 1:2
> G=diag(1,2)
> P=diag(2,2)
> (t(b)%*%G) / drop( sqrt( t(b)%*%P%*%b ) )
          [,1]      [,2]
[1,] 0.3162278 0.6324555

So far, so good. My problem is solved. However I found a little bit annoying
that R is not so "clever" as to realize that b'Pb can be
interpreted as a
scalar. I wonder :
would it be worth considering the implementation in R of
"recycling 1x1 matrix to scalars if appropriate"?
Just to leave the question on the ground...

Regards,

Andres
--
Andres Legarra
NEIKER
Apdo. 46
Vitoria-Gasteiz 01080 Spain
--





--
Andres Legarra Albizu
NEIKER
Apdo. 46
Vitoria-Gasteiz 01080 Spain
phone: +34 945 121323
fax: +34 945 281422
e-mail: alegarra at neiker.net
--

On Tue, 23 Aug 2005, Andres Legarra wrote:
> Hello,
> I am calculating this thing with vectors (b) and matrices (G,P):
> b'G/sqrt(b'Pb)
> where the denominator is a quadratic form and therefore always a scalar.
>
> In Scilab, it is quite simple:
> b'*G/sqrt(b'*P*b)
> However, in R, the denominator is an (1x1)matrix and R claims it is non
> conformable and I have to use drop() or as.numeric(). Like this:
>> b = 1:2
>> G=diag(1,2)
>> P=diag(2,2)
>> (t(b)%*%G) / drop( sqrt( t(b)%*%P%*%b ) )
>          [,1]      [,2]
> [1,] 0.3162278 0.6324555
>
> So far, so good. My problem is solved. However I found a little bit
annoying
> that R is not so "clever" as to realize that b'Pb can be
interpreted as a
> scalar. I wonder :
> would it be worth considering the implementation in R of
> "recycling 1x1 matrix to scalars if appropriate"?
> Just to leave the question on the ground...
This "clever"ness often leads to running code that the users did not 
intend to work that way and so give misleading answers.  In retrospect 
many of us think R/S should be less "clever" than it is, since
guessing
the minds of end-users is a dangerous pursuit.  In particular, as a recent 
posting on operator precedence shows, some users will have very different 
preconceptions from the developers, _and_ blame the developers for not 
having their perconceptions.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595