R help - Jul 2003 - Indexing with NA as FALSE??

Hi Folks,

Example:
t<-c(1,2,3,4,5,6,7,8,9)
u<-c(1,NA,3,NA,5,NA,7,NA,9)
t[u==5]
--> NA NA 5 NA NA

Now, if I could somehow set things so that "NA" was FALSE for
indexing, then
t[u==5]
--> 5

I know I can do it with
t[(u==5)&(!is.na(u))]
but in the situation I am dealing with this leads to massively
cumbersome, typo-prone and hard-to-read code.

Also, as an extra, it would be very useful if, for instance,
t[u==NA] --> 2 4 6 8
(I realise that working round this is less cumbersome, but even so).

What I'm really trying to work round is the "don't know" way
that R
handles NA. Reasonable in the logical sence, in that in the first
example R is saying "Can't tell whether u[2], u[4], u[6], u[8]
are equal to 5 or not, so will return a result which represents
this uncertainty", and in the second "You're telling me you
don't
know what value you want to match, so ... ".

Instead of that, since NA is one of the three values TRUE, FALSE, NA
of a logical, I'd like to be able to (a) treat NA as FALSE, (b) test
for a match between NA (as specified by me) and NA (as the value of
a logical variable).

Is there any non-cumbersome way to achieve this?

With thanks,
Ted

ted.harding at nessie.mcc.ac.uk wrote:
> I know I can do it with
> t[(u==5)&(!is.na(u))]
> but in the situation I am dealing with this leads to massively
> cumbersome, typo-prone and hard-to-read code.
  You could redefine '[' or '==', but that would lead to
massively
dangerous code. Anything could happen. Anyone who writes code that 
redefines such basic stuff may need their head examined.

  I think you are going to have to work round it with the !is.na(u) 
thing, but you could wrap it up in a function:

true4sure<-function(v){v & !is.na(v)}

then

 > t[true4sure(u==5)]
[1] 5

  although perhaps you could give it a less whimsical name....
> Also, as an extra, it would be very useful if, for instance,
> t[u==NA] --> 2 4 6 8
> (I realise that working round this is less cumbersome, but even so).
  Here is a way of doing that. It redefines '=='. It will break things 
that depend on NA's remaining NA's in comparisons. Do not use this code.
Do not even let it pollute your files. Consider it a dangerous virus:
> assign("==",function(a,b){a[is.na(a)]<-FALSE;
b[is.na(b)]<-FALSE; get("==","package:base")(a,b)})
  and then you get:
> c(1,2,3,NA,NA,NA) == c(1,NA,2,NA,NA,4)
[1]  TRUE FALSE FALSE  TRUE  TRUE FALSE
> Instead of that, since NA is one of the three values TRUE, FALSE, NA
> of a logical, I'd like to be able to (a) treat NA as FALSE, (b) test
> for a match between NA (as specified by me) and NA (as the value of
> a logical variable).
  Thats what it does. Of course it has a bug/feature in that NA is now 
== to FALSE.... But then you arent going to use that code.

  Safer would be to define a new binary operator:

 > assign("%=na%",function(a,b){a[is.na(a)]<-FALSE;
b[is.na(b)]<-FALSE;
  get("==","package:base")(a,b)})

  Then you can do:

 > c(1,2,3,NA,NA,NA) %=na% c(1,NA,2,NA,NA,4)
[1]  TRUE FALSE FALSE  TRUE  TRUE FALSE

  again this has the same NA==FALSE property.

  Here's a truth table for that operator:

 > outer(c(T,F,NA),c(T,F,NA),"%=na%")
       [,1]  [,2]  [,3]
[1,]  TRUE FALSE FALSE
[2,] FALSE  TRUE  TRUE
[3,] FALSE  TRUE  TRUE

  You just need to write an operator that returns TRUE on the diagonal 
only.... Easy modification of %=na% but its late on a Friday and I have 
a poker game to attend...

  Did I say not to use my code that redefines '=='? Well dont use it.
Ever.

Baz

On Fri, 11 Jul 2003 ted.harding at nessie.mcc.ac.uk wrote:
> Hi Folks,
>
> Example:
> t<-c(1,2,3,4,5,6,7,8,9)
> u<-c(1,NA,3,NA,5,NA,7,NA,9)
> t[u==5]
> --> NA NA 5 NA NA
>
> Now, if I could somehow set things so that "NA" was FALSE for
> indexing, then
> t[u==5]
> --> 5
t[u %in% 5]

> I know I can do it with
> t[(u==5)&(!is.na(u))]
> but in the situation I am dealing with this leads to massively
> cumbersome, typo-prone and hard-to-read code.
>
> Also, as an extra, it would be very useful if, for instance,
> t[u==NA] --> 2 4 6 8
t[ u %in% NA]


	-thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle

> Also, as an extra, it would be very useful if, for instance,
	> t[u==NA] --> 2 4 6 8
	> (I realise that working round this is less cumbersome, but even so).
	
What's wrong with writing t[is.na(u)]?

Or you could define
    eq <- function(x,y) {
	(is.na(x) & is.na(y)) | (!is.na(x) & !is.na(y) & x == y)
    }
and use
    t[eq(u,NA)]