[R] problem subsetting data frame with variable instead of constant

[Petr Savicky]

On Fri, Feb 10, 2012 at 08:15:39AM -0800, vaneet wrote:
> Hello,
> 
> I've encountered a very weird issue with the method subset(), or maybe this
> is something I don't know about said method that when you're subsetting
> based on the columns of a data frame you can only use constants (0.1, 2.3,
> 2.2) instead of variables?
> 
> Here's a look at my data frame called 'ea.cad.pwr':
> *>ea.ca.pwr[1:5,]
>    MAF   OR  POWER
> 1 0.02 0.01 0.9999
> 2 0.02 0.02 0.9998
> 3 0.02 0.03 0.9997
> 4 0.02 0.04 0.9995
> 5 0.02 0.05 0.9993*
> 
> Here's my subset lines which finds no rows:
> 
> *power1 = subset(ea.cad.pwr, MAF == maf1 & OR == odds)
> power2 = subset(ea.cad.pwr, MAF == maf2 & OR == odds)
> *
> Now when maf1 = 0.2 and odds = 1.2 it finds nothing.  I know for a fact that
> there's a row with these values:
> *> ea.cad.pwr[1430:1432,]
>      MAF   OR  POWER
> 1430 0.2 0.58 0.9996
> 1431 0.2 1.20 0.3092
> 1432 0.2 1.22 0.3914*
> 
> I have code working in a loop and each previous iteration the subset()
> function is working fine, but in this iteration some different lines are
> executed which are relevant to these variables, here they are:
> *
> maf1 = maf.adj - 0.01
> maf2 = maf.adj + 0.01*
> 
> Basically maf.adj is always a 2 decimal number (in this case = 0.21), and
> I'm computing the numbers around it by a difference of 0.01 (0.2,0.22) in
> case maf.adj isn't in the table.  maf.adj is read from another dataframe,
> when I use it to subset it always works fine but when I do this innocent
> subtraction for some reason it doesn't work.  If I rewrite statements like
> this it works:
> 
> *power1 = subset(ea.cad.pwr, MAF == 0.2 & OR == odds)
> power2 = subset(ea.cad.pwr, MAF == 0.22 & OR == odds)
> *
> 
> Even if I write this first:
> 
> maf1 = 0.2
> 
> Then:
> 
> power1 = subset(ea.cad.pwr, MAF == maf1 & OR == odds)
> 
> It works as well! That's what's really confusing, how can this subtraction
> mess everything up?  Please help if you can..thank you!

Hi.

This may be a rounding problem. Try

  0.3 - 0.1 == 0.2

  [1] FALSE

Explicit rounding to a not too large number of decimal
digits can help.

  round(0.3 - 0.1, digits=7) == 0.2

  [1] TRUE


See also FAQ 7.31 or http://rwiki.sciviews.org/doku.php?id=misc:r_accuracy

Hope this helps.

Petr Savicky.