R devel - Oct 2003 - zapsmall with all values small (PR#4554)

Hi

If all values are small, zapsmall does not zap.

For example ...

 > zapsmall(1e-24)
[1] 1e-24
 > zapsmall(1e-24, digits=3)
[1] 1e-24

Problem appears to be in calculation of digits ...

     mx <- max(abs(x[!ina]))
     digits = if (mx > 0)
                max(0, digits - log10(mx))
              else
                digits

If all values are small, mx is very small, log10(mx) is "large" but 
negative, digits - log10(mx) is very large, so no rounding occurs.

Paul
--please do not edit the information below--

Version:
  platform = i686-pc-linux-gnu
  arch = i686
  os = linux-gnu
  system = i686, linux-gnu
  status = Under development (unstable)
  major = 1
  minor = 9.0
  year = 2003
  month = 10
  day = 12
  language = R

Search Path:
  .GlobalEnv, package:gridBase, package:methods, package:ctest, 
package:mva, package:modreg, package:nls, package:ts, package:lattice, 
package:grid, package:graphicsQC, Autoloads, package:base

-- 
Dr Paul Murrell
Department of Statistics
The University of Auckland
Private Bag 92019
Auckland
New Zealand
64 9 3737599 x85392
paul@stat.auckland.ac.nz
http://www.stat.auckland.ac.nz/~paul/

>>>>> "Paul" == p murrell
<p.murrell@auckland.ac.nz>
>>>>>     on Mon, 13 Oct 2003 01:39:21 +0200 (MET DST) writes:
    Paul> Hi If all values are small, zapsmall does not zap.

    Paul> For example ...

    >> zapsmall(1e-24)
    Paul> [1] 1e-24
    >> zapsmall(1e-24, digits=3)
    Paul> [1] 1e-24


    Paul> Problem appears to be in calculation of digits ...

    Paul>      mx <- max(abs(x[!ina])) digits = if (mx > 0)
    Paul> max(0, digits - log10(mx)) else digits

    Paul> If all values are small, mx is very small, log10(mx)
    Paul> is "large" but negative, digits - log10(mx) is very
    Paul> large, so no rounding occurs.

This has very much been the intent of zapsmall AFAI have
"forever" known (though the documentation unfortunately does not
tell clearly so) :

The "small"  in "zapsmall(x)" is something  *relative* to
the other
numbers in `x'.

If you know in advance that 1e-24 is small (it isn't, compared
to 1e-22 or even 1e-50), then use round() with an explicit
digits argument.

Hence, the bug is only in the not-so-clear documentation.

Martin