R help - Nov 2012 - [lattice] format and rotation of strip text

Thanks to the lattice gurus on this list, and having reference to the
excellent open-access Sarkar 2008

ISBN 978-0-387-75968-5
e-ISBN 978-0-387-75969-2
http://dx.doi.org/10.1007/978-0-387-75969-2

I now know how to label lattice panels by variable value: see thread
starting @

https://stat.ethz.ch/pipermail/r-help/2012-November/329450.html

(and demonstrated below). This allows me to use lattice::levelplot() to
display atmospheric data (gas concentrations) over a 3D space with
dimensions longitude, latitude, and (vertical) level ... but I would
like to fix a few things. Hopefully the following is a sufficiently
"small, self-contained example" (though it has 2 side questions
injected--answers to those are appreciated also):

# start example
library(reshape2)
library(lattice)

lon=11
lat=7
lev=5
len=lon*lat*lev
array.3d <- array(data=c(1:len), dim=c(lat, lon, lev))

# Rewrite the array values "more spatially," i.e., row-wise from
# bottom left. If there's a more-R-ish way to fill this array
# as specified, please let me know: I know 'for' loops are deprecated
# in R.

i=1
for (z in 1:lev) {
  for (x in lat:1) {
    for (y in 1:lon) {
      array.3d[x,y,z]=i ; i=i+1
    }
  }
}

# produces (with rows=latitudes and cols=longitudes)
array.3d[,,1]
array.3d[,,lev]

# convert data=array.3d to dataframe with reshape2::melt
array.3d.df <- melt(array.3d,
varnames=c("lat","lon","lev"),
value.name="conc")
head(array.3d.df)
tail(array.3d.df)

# make level values {longer, "more realistic"}

array.3d.df$lev <- array.3d.df$lev + 0.12345
head(array.3d.df)
tail(array.3d.df)

# plot "appropriately" for atmospheric data where lev=pressure: use
# * lattice::levelplot
# * one column, since atmospheric levels stack vertically
# * rev(lev), since layers closer to ground level have higher pressure
levelplot(
  conc ~ lon * lat | rev(lev), data=array.3d.df,
  layout=c(1,lev),  # show levels stacked in 1 vertical column
  strip=FALSE,      # this suppresses printing strips atop packets
  strip.left=strip.custom(
    strip.levels=TRUE,  # print level values
    strip.names=FALSE   # don't print name of level
variable="rev(lev)"
  )
)
# end example

Note that the (colored) 'strip' for each panel in the lattice has

- the corresponding layer value printed inside curly brackets, e.g.,
  '{1.12345}'

- the layer value printed in full

- the layer value rotated 90? CCW (like the y-axis label)

I would prefer to have

+ the layer value *not* printed inside curly brackets

+ the layer value *not* rotated 90? CCW (i.e., to print the layer value
  like the x-axis label)

+ the layer value truncated or rounded to some significant digits,
  e.g., '1.1'

I suspect this can be done with strip.custom, but am not seeing how;
please enlighten!

Your assistance is appreciated, Tom Roche <Tom_Roche at pobox.com>

On Nov 18, 2012, at 12:42 PM, Tom Roche wrote:
>
> Thanks to the lattice gurus on this list, and having reference to the
> excellent open-access Sarkar 2008
>
> ISBN 978-0-387-75968-5
> e-ISBN 978-0-387-75969-2
> http://dx.doi.org/10.1007/978-0-387-75969-2
>
> I now know how to label lattice panels by variable value: see thread
> starting @
>
> https://stat.ethz.ch/pipermail/r-help/2012-November/329450.html
>
> (and demonstrated below). This allows me to use lattice::levelplot()  
> to
> display atmospheric data (gas concentrations) over a 3D space with
> dimensions longitude, latitude, and (vertical) level ... but I would
> like to fix a few things. Hopefully the following is a sufficiently
> "small, self-contained example" (though it has 2 side questions
> injected--answers to those are appreciated also):
>
> # start example
> library(reshape2)
> library(lattice)
>
> lon=11
> lat=7
> lev=5
> len=lon*lat*lev
> array.3d <- array(data=c(1:len), dim=c(lat, lon, lev))
>
> # Rewrite the array values "more spatially," i.e., row-wise from
> # bottom left. If there's a more-R-ish way to fill this array
> # as specified, please let me know: I know 'for' loops are
deprecated
> # in R.
>
> i=1
> for (z in 1:lev) {
>  for (x in lat:1) {
>    for (y in 1:lon) {
>      array.3d[x,y,z]=i ; i=i+1
>    }
>  }
> }
>
> # produces (with rows=latitudes and cols=longitudes)
> array.3d[,,1]
> array.3d[,,lev]
>
> # convert data=array.3d to dataframe with reshape2::melt
> array.3d.df <- melt(array.3d,
varnames=c("lat","lon","lev"),
> value.name="conc")
> head(array.3d.df)
> tail(array.3d.df)
>
> # make level values {longer, "more realistic"}
>
> array.3d.df$lev <- array.3d.df$lev + 0.12345
> head(array.3d.df)
> tail(array.3d.df)
>
> # plot "appropriately" for atmospheric data where lev=pressure:
use
> # * lattice::levelplot
> # * one column, since atmospheric levels stack vertically
> # * rev(lev), since layers closer to ground level have higher pressure
> levelplot(
>  conc ~ lon * lat | rev(lev), data=array.3d.df,
>  layout=c(1,lev),  # show levels stacked in 1 vertical column
>  strip=FALSE,      # this suppresses printing strips atop packets
>  strip.left=strip.custom(
>    strip.levels=TRUE,  # print level values
>    strip.names=FALSE   # don't print name of level
variable="rev(lev)"
>  )
> )
> # end example
>
> Note that the (colored) 'strip' for each panel in the lattice has
>
> - the corresponding layer value printed inside curly brackets, e.g.,
>  '{1.12345}'
>
> - the layer value printed in full
>
> - the layer value rotated 90? CCW (like the y-axis label)
>
> I would prefer to have
>
> + the layer value *not* printed inside curly brackets
>
> + the layer value *not* rotated 90? CCW (i.e., to print the layer  
> value
>  like the x-axis label)
>
> + the layer value truncated or rounded to some significant digits,
>  e.g., '1.1'
levelplot(
  conc ~ lon * lat | rev(lev), data=array.3d.df,
  layout=c(1,lev),   levs=as.character(round(array.3d.df[['lev']], 1)),
  strip=FALSE,
  strip.left=strip.custom(
    factor.levels=as.character(round(array.3d.df[['lev']], 3)),
    strip.levels=TRUE,
    horizontal=TRUE,
    strip.names=FALSE ,
    par.strip.text=list( cex=0.5)
  )
)
>
> I suspect this can be done with strip.custom, but am not seeing how;
> please enlighten!
Your example does not do a very good job of testing the levels  
assignments since they are all the same.

-- 

David Winsemius, MD
Alameda, CA, USA

https://stat.ethz.ch/pipermail/r-help/2012-November/329479.html
>> Hopefully [sufficiently] "small, self-contained example":
mailquotes omitted from "start example" to "end example" to
ease rerunning the following code:

# start example
library(reshape2)
library(lattice)

lon=11
lat=7
lev=5
len=lon*lat*lev
array.3d <- array(data=c(1:len), dim=c(lat, lon, lev))

# Rewrite the array values "more spatially," i.e., row-wise from
# bottom left. If there's a more-R-ish way to fill this array as
# desired, please let me know: I know 'for' loops are deprecated
# in R.

i=1
for (z in 1:lev) {
  for (x in lat:1) {
    for (y in 1:lon) {
      array.3d[x,y,z]=i ; i=i+1
    }
  }
}

# produces (with rows=latitudes and cols=longitudes)
array.3d[,,1]
array.3d[,,lev]

# convert data=array.3d to dataframe with reshape2::melt
array.3d.df <- melt(array.3d,
varnames=c("lat","lon","lev"),
value.name="conc")
head(array.3d.df)
tail(array.3d.df)

# make level values {longer, "more realistic"}

array.3d.df$lev <- array.3d.df$lev + 0.12345 # truncated below, and ...
# ... below note output from these
head(array.3d.df)
tail(array.3d.df)

# plot "appropriately" for atmospheric data where lev=pressure: use
# * lattice::levelplot
# * one column, since atmospheric levels stack vertically
# * rev(lev), since layers closer to ground level have higher pressure
levelplot(
  conc ~ lon * lat | rev(lev), data=array.3d.df,
  layout=c(1,lev),  # show levels stacked in 1 vertical column
  strip=FALSE,      # this suppresses printing strips atop packets
  strip.left=strip.custom(
    strip.levels=TRUE,  # print level values
    strip.names=FALSE   # don't print name of level
variable="rev(lev)"
  )
)
# end example
>> Note that the (colored) 'strip' for each panel in the lattice
has
>> - the corresponding layer value printed inside curly brackets, e.g.,
>>   '{1.12345}'
>> - the layer value printed in full
>> - the layer value rotated 90? CCW (like the y-axis label)
>> I would prefer to have
>> + the layer value *not* printed inside curly brackets
>> + the layer value *not* rotated 90? CCW (i.e., to print the layer
>>   value like the x-axis label)
>> + the layer value truncated or rounded to some significant digits,
>>   e.g., '1.1'
https://stat.ethz.ch/pipermail/r-help/2012-November/329505.html
> levelplot(
>   conc ~ lon * lat | rev(lev), data=array.3d.df,
>   layout=c(1,lev),   levs=as.character(round(array.3d.df[['lev']],
1)),
>   strip=FALSE,
>   strip.left=strip.custom(
>     factor.levels=as.character(round(array.3d.df[['lev']], 3)),
>     strip.levels=TRUE,
>     horizontal=TRUE,
>     strip.names=FALSE ,
>     par.strip.text=list( cex=0.5)
>   )
> )
...
> Your example does not do a very good job of testing the levels
> assignments since they are all the same.
Actually, that claim is false, as can be demonstrated in 2 ways:

1. (simple) Note column=lev in output from

head(array.3d.df)
tail(array.3d.df)

    The levels are not "all the same."

2. (more complex, but clearly demonstrates a flaw in the suggestion)
   Shorten the level values, then comment out the 'factor.levels'
   argument to strip.custom (above), and plot:

# shorten level values by 2 digits, for bug demonstration
array.3d.df$lev <- array.3d.df$lev - 0.00045 
# note different output from these in column=lev
head(array.3d.df)
tail(array.3d.df)

# comment out for debugging
levelplot(
  conc ~ lon * lat | rev(lev), data=array.3d.df, layout=c(1,lev),
  levs=as.character(round(array.3d.df[['lev']], 1)),
  strip=FALSE,
  strip.left=strip.custom(
#    factor.levels=as.character(round(array.3d.df[['lev']], 3)),
    strip.levels=TRUE,
    horizontal=TRUE,
    strip.names=FALSE ,
    par.strip.text=list(cex=0.5)
  )
)

   Note that,

-  without the line above commented out, strip values are (correctly)
   all 1.123

+  with the line above commented out, strip values are (correctly)
   1.123 .. 5.123

Am I missing something? If not, how to round() or signif() the values
obtained from array.3d.df$lev and displayed in the strips?

Your assistance is appreciated, Tom Roche <Tom_Roche at pobox.com>