[R] dealing with NA in readBin() and writeBin()

Mike Miller mbmiller+l at gmail.com
Sun Jan 4 23:40:13 CET 2015


On Sun, 4 Jan 2015, Duncan Murdoch wrote:

> On 04/01/2015 5:13 PM, Mike Miller wrote:
>> The help doc for readBin writeBin tells me this:
>>
>> Handling R's missing and special (Inf, -Inf and NaN) values is discussed
>> in the ‘R Data Import/Export’ manual.
>>
>> So I go here:
>>
>> http://cran.r-project.org/doc/manuals/r-release/R-data.html#Special-values
>>
>> Unfortunately, I don't really understand that.  Suppose I am using
>> single-byte integers and I want 255 (binary 11111111) to be translated to
>> NA.  Is it possible to do that?  Of course I could always do something
>> like this:
>>
>> X[ X==255 ] <- NA
>>
>> The problem with that is that I want to process the data on the fly,
>> dividing the integer to produce a double in the range from 0 to 2:
>>
>> X <- readBin( file, what="integer", n=N, size=1, signed=FALSE)/127
>
> Why?  Why not do it in three steps, i.e.
>
> X <- readBin( file, what="integer", n=N, size=1, signed=FALSE)
> X[ X==255 ] <- NA
> X <- X/127
>
> If you are worried about the extra typing, then write a function to 
> handle all three steps.

The thing I was concerned about is the memory usage, not the typing, 
because everything will be scripted.  But maybe memory isn't an issue and 
I never have to hold two copies in memory simultaneously.  There will be 
about 50 million elements, typically.

I think in terms of processing numbers that are streaming into memory, but 
that might not be what R is doing.  For example, with scan() and 
na.strings="NA", I picture it changing strings to NA as they are read, it 
might load the whole file as character, then do all the work with things 
like what=numeric() and na.strings="NA" after the fact.  Maybe that 
doesn't impose an extra memory burden.


>> It looks like this still works:
>>
>> X[ X==255/127 ] <- NA
>
> I suspect that would work on all current platforms, but I wouldn't trust 
> it.  Don't use == on floating point values unless you know they are 
> fractions with 2^n in the denominator.

Good point about platforms.  I was concerned about the use of ==, and 
you've convinced me it is not trustworthy.

Thanks very much.

Mike


More information about the R-help mailing list