2:39 p.m.
Dear all,
I know this is not the gnuplot forum but I hope that somebody knows how
to solve this using gretl + gnuplot (v. 5 patchlevel 5).
I want to plot a stacked barplot over time by providing the matrix 'mat'
(rows=Time, cols=variables) and a vector 'xmat'. xmat is based on the
$obsdate command holding daily dates in the format YYYYMMDD. My
objective is to handle date strings on the x-axis correctly. But this
drives me nuts...
As you will see, for the moment I deactivated everything related to time
and its formatting. I also took Jack's function for plotting the
historical decomposition based on the SVAR package as a role model but
somehow this doesn't work for me.
I attached the function plus an example which results in the error:
<gnuplot>
plot '-' using 1:($2), '-' using 1:($2)
^
"/home/at/.gretl/gpttmp.DjR2By", line 25: Too many columns in using
specification
*** error in function stackplot, line 70
</gnuplot>
Btw, do we have a function by retrieving the seconds since the Unix
epoch (1970-01-01, 00:00 UTC). I am just curious as gnuplot can handle
this time format
(http://www.phy.ntnu.edu.tw/demolab/html.php?html=doc/gnuplot-doc/html/tim...).
Thanks in advance,
Artur
1:05 a.m.
On Tue, 5 Dec 2017, Artur Tarassow wrote:
Dear all,
I know this is not the gnuplot forum but I hope that somebody knows how to
solve this using gretl + gnuplot (v. 5 patchlevel 5).
I want to plot a stacked barplot over time by providing the matrix 'mat'
(rows=Time, cols=variables) and a vector 'xmat'. xmat is based on the
$obsdate command holding daily dates in the format YYYYMMDD. My objective is
to handle date strings on the x-axis correctly. But this drives me nuts...
As you will see, for the moment I deactivated everything related to time and
its formatting. I also took Jack's function for plotting the historical
decomposition based on the SVAR package as a role model but somehow this
doesn't work for me.
I attached the function plus an example which results in the error:
<gnuplot>
plot '-' using 1:($2), '-' using 1:($2)
^
"/home/at/.gretl/gpttmp.DjR2By", line 25: Too many columns in using
specification
You're only allowed one data column in a gnuplot histogram plot
clause.
I'm sure that more elegant variants are possible, but here's an
example that does roughly what you want (I think):
<hansl>
set verbose off
function void stackplot (matrix mat,
matrix xmat[null] "add if you want tp plot vs. xmat-values",
int nstep[1::4] "Print only n-th value on xaxis",
string fname)
tmpfile = sprintf("%s/stackedbar.gp", $dotdir)
outfile @tmpfile --write
printf "set encoding utf8\n"
printf "set nokey\n"
printf "set style data histogram \n"
printf "set style histogram rowstacked\n"
printf "set style fill solid border -1\n"
printf "set style fill solid 0.35\n"
printf "set xtics border in scale 0,0 nomirror rotate by -45 autojustify\n"
# write data block
printf "$data <<EOF\n"
loop i=1..rows(mat) -q
loop j=1..cols(mat) -q
printf "%g ", mat[i,j]
endloop
if exists(xmat)
if i % nstep == 0
printf "%g", xmat[i]
else
printf "-1"
endif
endif
printf "\n"
endloop
printf "EOF\n"
# write plot specification
printf "plot \\\n"
loop j=1..cols(mat) -q
if j == 1 && exists(xmat)
k = cols(mat) + 1
printf "$data using 1:xtic($%d < 0. ? \"\" :
stringcolumn(%d))", k, k
else
printf "$data using %d", j
endif
printf(j == cols(mat) ? "\n" : ", \\\n")
endloop
outfile --close
gnuplot --input="@tmpfile" --output=@fname
end function
open data9-7 -q
smpl 1975:1 1977:4
series date = (int($obsdate/100))/100
series QNC2 = QNC * 0.5
matrix M = {QNC}~{QNC2}
stackplot(M, {date}, 2, "display")
</hansl>
Allin
7:40 a.m.
On Tue, 5 Dec 2017, Allin Cottrell wrote:
On Tue, 5 Dec 2017, Artur Tarassow wrote:
> Dear all,
[...]
>
> I want to plot a stacked barplot over time by providing the matrix 'mat'
> (rows=Time, cols=variables) and a vector 'xmat'. xmat is based on the
> $obsdate command holding daily dates in the format YYYYMMDD. My objective
> is to handle date strings on the x-axis correctly. But this drives me
> nuts...
[...]
I'm sure that more elegant variants are possible, but here's
an example that
does roughly what you want (I think):
[...]
I wasn't aware that gnuplot has the bash-style "<<EOF" thingie. Very
nice.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
6:11 p.m.
On Tue, 5 Dec 2017, Artur Tarassow wrote:
I want to plot a stacked barplot over time by providing the matrix
'mat' (rows=Time, cols=variables) and a vector 'xmat'. xmat is based
on the $obsdate command holding daily dates in the format YYYYMMDD.
My objective is to handle date strings on the x-axis correctly. But
this drives me nuts...
Here's a somewhat improved version of my suggestion. As it stands it's
adapted to quarterly data. It would be straightforward to generalize
it to cover monthly data, daily would require a little more work.
In this version I've simplified things a little by assuming that
Artur's "xmat" (time variable) is not an optional extra but is the
last column of the matrix argument to stackplot().
<hansl>
set verbose off
function void stackplot (matrix mat,
int nstep[1::4] "Print only n-th value on xaxis",
string fname)
k = cols(mat)
tmpfile = sprintf("%s/stackedbar.gp", $dotdir)
outfile @tmpfile --write
printf "set encoding utf8\n"
printf "set nokey\n"
printf "set style data histogram \n"
printf "set style histogram rowstacked\n"
printf "set style fill solid border -1\n"
printf "set style fill solid 0.35\n"
printf "set xtics nomirror rotate by -45\n"
# write data block
printf "$data << EOF\n"
loop i=1..rows(mat) -q
loop j=1..k-1 -q
printf "%g ", mat[i,j]
endloop
cond = (i == 1 || (i+1) % nstep == 0)
printf "%g\n", cond ? mat[i,k] : -1
endloop
printf "EOF\n"
# define a gnuplot function
printf "# start literal lines\n"
printf "qdate(x)=sprintf(\"%%gQ%%g\",floor(x),10*(x-floor(x)))\n"
printf "# end literal lines\n"
# write plot specification
printf "plot \\\n"
loop j=1..k-1 -q
if j == 1
printf "$data using 1:xtic($%d < 0. ? \"\" : qdate($%d))",
k, k
else
printf "$data using %d", j
endif
printf "%s\n", j == k-1 ? "" : ",\"
endloop
outfile --close
gnuplot --input="@tmpfile" --output="@fname"
end function
open data9-7 -q
smpl 1975:1 1980:4
# quarterly data
series date = $obsmajor + $obsminor/10
stackplot({QNC}~{QNC/2}~{date}, 2, "display")
</hansl>
Allin
6:53 p.m.
Thank you very much, Allin. I am just busy with preparing lectures but I
am going to have a look at it tomorrow.
Best,
Artur
Am 06.12.2017 um 19:11 schrieb Allin Cottrell:
On Tue, 5 Dec 2017, Artur Tarassow wrote:
> I want to plot a stacked barplot over time by providing the matrix
> 'mat' (rows=Time, cols=variables) and a vector 'xmat'. xmat is based
> on the $obsdate command holding daily dates in the format YYYYMMDD. My
> objective is to handle date strings on the x-axis correctly. But this
> drives me nuts...
Here's a somewhat improved version of my suggestion. As it stands it's
adapted to quarterly data. It would be straightforward to generalize it
to cover monthly data, daily would require a little more work.
In this version I've simplified things a little by assuming that Artur's
"xmat" (time variable) is not an optional extra but is the last column
of the matrix argument to stackplot().
<hansl>
set verbose off
function void stackplot (matrix mat,
int nstep[1::4] "Print only n-th value on xaxis",
string fname)
k = cols(mat)
tmpfile = sprintf("%s/stackedbar.gp", $dotdir)
outfile @tmpfile --write
printf "set encoding utf8\n"
printf "set nokey\n"
printf "set style data histogram \n"
printf "set style histogram rowstacked\n"
printf "set style fill solid border -1\n"
printf "set style fill solid 0.35\n"
printf "set xtics nomirror rotate by -45\n"
# write data block
printf "$data << EOF\n"
loop i=1..rows(mat) -q
loop j=1..k-1 -q
printf "%g ", mat[i,j]
endloop
cond = (i == 1 || (i+1) % nstep == 0)
printf "%g\n", cond ? mat[i,k] : -1
endloop
printf "EOF\n"
# define a gnuplot function
printf "# start literal lines\n"
printf "qdate(x)=sprintf(\"%%gQ%%g\",floor(x),10*(x-floor(x)))\n"
printf "# end literal lines\n"
# write plot specification
printf "plot \\\n"
loop j=1..k-1 -q
if j == 1
printf "$data using 1:xtic($%d < 0. ? \"\" : qdate($%d))",
k, k
else
printf "$data using %d", j
endif
printf "%s\n", j == k-1 ? "" : ",\"
endloop
outfile --close
gnuplot --input="@tmpfile" --output="@fname"
end function
open data9-7 -q
smpl 1975:1 1980:4
# quarterly data
series date = $obsmajor + $obsminor/10
stackplot({QNC}~{QNC/2}~{date}, 2, "display")
</hansl>
3:32 p.m.
Dear Allin,
thanks again for your examples. These are nice illustrations and I
wasn't aware of some of the capabilities of gnuplot.
In order to handle different time frequencies more flexible, I thought
it could of an advantage to exploit the $obsdate series which actually
contains all necessary information. In the attached example, I transform
the int values of $obsdate into an array of strings and use these string
values (separating year, month, day) as xtics in a flexible manner. This
also allows the user to define different options on how to present
dates, as you will see below.
Btw, there is a weird thing for large matrices with my obs. as used
below. In order to generate a proper graph one needs to write:
<hansl>
printf "set style fill solid border -2\n" # "border -1" doesn't
work for
long data!
</hansl>
Best,
Artur
<hansl>
set verbose off
function void stackplot1AT (matrix mat,
series obsdate[null] "$obsdate",
int dateform[0:4:1] "Output format of xtics [see asdate()]",
int nstep[1::4] "Print only n-th value on xaxis",
string fname,
int height[240::480], int width[240::640]))
if exists(obsdate)
# re-define obsdate as array of strings
strings S = s2a(obsdate)
strings Date = asdate(S,dateform)
endif
tmpfile = sprintf("%s/stackedbar.gp", $dotdir)
outfile @tmpfile --write
printf "set encoding utf8\n"
printf "set terminal pngcairo size %d,%d enhanced font
'Verdana,12'\n", width,height
printf "set nokey\n"
printf "set style data histogram \n"
printf "set style histogram rowstacked\n"
printf "set style fill solid border -2\n" # "border -1"
doesn't
work for long data!
printf "set style fill solid 0.35\n"
printf "set xtics border in scale 0,0 nomirror rotate by -45
autojustify\n"
# write data block
printf "$data <<EOF\n"
loop i=1..rows(mat) -q
loop j=1..cols(mat) -q
printf "%g ", mat[i,j]
endloop
if exists(obsdate)
if i % nstep == 0
printf "%s", Date[i]
else
printf "-1"
endif
endif
printf "\n"
endloop
printf "EOF\n"
# write plot specification
printf "plot \\\n"
loop j=1..cols(mat) -q
if j == 1 && exists(obsdate)
k = cols(mat) + 1
printf "$data using 1:xtic($%d < 0. ? \"\" :
stringcolumn(%d))", k, k
else
printf "$data using %d", j
endif
printf(j == cols(mat) ? "\n" : ", \\\n")
endloop
outfile --close
gnuplot --input="@tmpfile" --output=@fname
end function
function strings s2a (series y)
# put numerical values into an array of strings
n = nobs(y)
strings S=array(n)
loop i=1..n -q
sprintf s "%d", y[i]
S[i] = "@s"
endloop
return S
end function
function strings asdate (strings S "stringified $obsdate array",
int dform[0:4:1] "0=Y/M/D, 1=y/M/D, 2=Y/M, 3=y/M, 4=M/D")
n=nelem(S)
strings D = array(n)
loop i=1..n -q
string s = S[i]
string y = substr(s,1,4) # year
string m = substr(s,5,6) # month
string d = substr(s,7,8) # day
if dform==0
D[i] = sprintf("%s/%s/%s",y,m,d)
elif dform==1
y = substr(y,3,4)
D[i] = sprintf("%s/%s/%s",y,m,d)
elif dform==2
D[i] = sprintf("%s/%s",y,m)
elif dform==3
y = substr(y,3,4)
D[i] = sprintf("%s/%s",y,m)
elif dform==4
D[i] = sprintf("%s/%s",m,d)
endif
endloop
return D
end function
# Example using daily data
open djclose -q
rename djclose dj
series date = $obsdate
series dj2 = dj
matrix M = {dj}~{dj2}
dateform = 4 # option for xtics format
xskip = 150
stackplot1AT(M, date, 4, xskip, "display", 320,720)
</hansl>
2:30 a.m.
On Thu, 7 Dec 2017, Artur Tarassow wrote:
Dear Allin,
thanks again for your examples. These are nice illustrations and I
wasn't aware of some of the capabilities of gnuplot.
In order to handle different time frequencies more flexible, I
thought it could of an advantage to exploit the $obsdate series
which actually contains all necessary information. In the attached
example, I transform the int values of $obsdate into an array of
strings and use these string values (separating year, month, day)
as xtics in a flexible manner. This also allows the user to define
different options on how to present dates, as you will see below.
Yes, that's nice, I agree.
Btw, there is a weird thing for large matrices with my obs. as
used below. In order to generate a proper graph one needs to
write:
<hansl>
printf "set style fill solid border -2\n"
# "border -1" doesn't work for long data!
</hansl>
Not really so weird. The number that (can) follow "border" in this
context is a line-type. Line type -1 is a solid black line. So with
"border -1" and the daily djclose data you're requesting that each
of the 2528 boxes is bordered by a black line. That produces a
totally dense black mass with no pixels left to show any colors.
Line type -2 doesn't exist, so (by courtesy, it seems) if you say
"border -2" that's equivalent to saying "noborder", in which case
the selected colors for the boxes are visible.
Allin
9:34 a.m.
Yes, that's nice, I agree.
Glad you like it ;-)
> Btw, there is a weird thing for large matrices with my obs. as
used
> below. In order to generate a proper graph one needs to write:
>
> <hansl>
> printf "set style fill solid border -2\n"
> # "border -1" doesn't work for long data!
> </hansl>
Not really so weird. The number that (can) follow "border" in this
context is a line-type. Line type -1 is a solid black line. So with
"border -1" and the daily djclose data you're requesting that each of
the 2528 boxes is bordered by a black line. That produces a totally
dense black mass with no pixels left to show any colors.
Line type -2 doesn't exist, so (by courtesy, it seems) if you say
"border -2" that's equivalent to saying "noborder", in which case
the
selected colors for the boxes are visible.
Thank you for this clarification. Then
it makes sense to check the
number of boxes and set the option for border accordingly.
Best,
Artur
2568
days inactive
2573
days old
7 comments
4 participants
participants (4)
-
Allin Cottrell -
Artur T. -
Artur Tarassow
-
Riccardo (Jack) Lucchetti