R devel - Feb 2005 - Incorrect behavior for ordering timepoints in "reshape" (PR#7669)

Full_Name: Dav Clark
Version: 2.0.1
OS: OS X 10.3
Submission from: (NULL) (128.122.87.35)


When the timepoints that reshape uses (in direction="long") are
negative or
fractional, the time label is assigned incorrectly.  It is easier to give an
example than to describe the problem abstractly:

Assume you have a data.frame header with values related to peri-stimulus time
like this:

"HRF -5" "HRF -2.5" "HRF 0" "HRF 2.5"
... "HRF 10"

And you give reshape a split argument of a space " ".

Then the labels will be assigned strangely, based on alphabetical ordering.  So
the above list order maps to:

-2.5, -5, 0, 10, ... 2.5

Items under the "HRF -5" column in wide format recieve a -2.5 label,
items under
"HRF 2.5" receive a label of 10, and so on.

Somewhere, the time labels are being used before conversion to numbers.  But,
reshape returns an error if it is not possible to convert the timepoints to
numeric!  So obviously, more functionality could be provided, or at least the
documentation should reflect the current shortfall.

For completeness, here is a minimal example demonstrating the bug:

df <- data.frame(id="S1", V1="from -2", V2="from
-1")
names(df)[2:3] <- c("vals.-2", "vals.-1")
df
reshape(df, direction="long", varying=2:3)

Thanks!
Dav

davclark@nyu.edu writes:
> Full_Name: Dav Clark
> Version: 2.0.1
> OS: OS X 10.3
> Submission from: (NULL) (128.122.87.35)
> 
> 
> When the timepoints that reshape uses (in direction="long") are
negative or
> fractional, the time label is assigned incorrectly.  It is easier to give
an
> example than to describe the problem abstractly:
> 
> Assume you have a data.frame header with values related to peri-stimulus
time
> like this:
> 
> "HRF -5" "HRF -2.5" "HRF 0" "HRF
2.5" ... "HRF 10"
> 
> And you give reshape a split argument of a space " ".
> 
> Then the labels will be assigned strangely, based on alphabetical ordering.
So
> the above list order maps to:
> 
> -2.5, -5, 0, 10, ... 2.5
> 
> Items under the "HRF -5" column in wide format recieve a -2.5
label, items under
> "HRF 2.5" receive a label of 10, and so on.
> 
> Somewhere, the time labels are being used before conversion to numbers. 
But,
> reshape returns an error if it is not possible to convert the timepoints to
> numeric!  So obviously, more functionality could be provided, or at least
the
> documentation should reflect the current shortfall.
> 
> For completeness, here is a minimal example demonstrating the bug:
> 
> df <- data.frame(id="S1", V1="from -2",
V2="from -1")
> names(df)[2:3] <- c("vals.-2", "vals.-1")
> df
> reshape(df, direction="long", varying=2:3)
Hmm, this looks messed up even without the negatives. The guess()
function inside reshape always sorts before converting to numeric, so
you get the 1 10 11 2 3 4 5 6 7 8 9 effect, but what is worse: the
sorting decouples the values from the variable names, as demonstrated
by modifying your example slightly
> reshape(df, direction="long", varying=3:2)
      id time    vals
S1.-1 S1   -1 from -1
S1.-2 S1   -2 from -2

I'm not at all sure I understand what was supposed to happen here,
perhaps the sort in

    varying <- unique(nn[, 1])
    times <- sort(unique(nn[, 2]))

is a thinko? Over to Thomas, I think.

-- 
   O__  ---- Peter Dalgaard             Blegdamsvej 3  
  c/ /'_ --- Dept. of Biostatistics     2200 Cph. N   
 (*) \(*) -- University of Copenhagen   Denmark      Ph: (+45) 35327918
~~~~~~~~~~ - (p.dalgaard@biostat.ku.dk)             FAX: (+45) 35327907