Re: [Gretl-users] First time to plot a series after installation of gretl
On Wed, 14 Jul 2010, peter wrote:
I tried wgnuplot, found it loaded and plotted instantly. Rebooted
and
found no difference.
I tried GRETL again, tested for graphing from console vs right-click on
a series. I thought I perceived a faster time from the console -- but
if there was, it was marginal.
I noticed this graph delay in my class last spring, but after the first
few sessions the students ignored it. All the machines in the lab are
older PCs, low memory.
Thanks, Peter. My experience matches yours. I have occasionally
seen quite slow time-to-plot but I find it's not reproducible.
This evening I tried an experiment on MS Windows Vista, after
putting into gretl some timing code to measure the various phases
of the on-screen plotting operation, which are basically three:
(i) prepare the source file to send to gnuplot; (ii) call gnuplot
to generate a PNG file; and (iii) process the PNG file and get its
contents into a GTK window. I thought that if the plot took a long
time to appear I would see where the problem was.
I took what I thought was the worst-case scenario: boot Windows,
remove gretl, install the current snapshot, start gretl, and try a
plot. I did, however, let Windows "settle" a little after booting
(opened Firefox, waited several seconds until it became properly
responsive, visited the gretl site and downloaded the Windows
snapshot, closed Firefox.)
Anyway, here are my timings (note that they're not cumulative,
each figure represents seconds elapsed since the last
way-station):
# first plot (single time-series)
calling gnuplot()
execing wgnuplot.exe: 0.016
wgnuplot.exe done: 0.717
gnuplot() returned 0: 0
gnuplot_show_png starting: 0
done read_plotspec_from_file: 0
get_png_bounds_info(): OK
calling render_pngfile: 0.016
done render_pngfile: 0.062
# second plot (single time-series)
calling gnuplot()
execing wgnuplot.exe: 0
wgnuplot.exe done: 0.343
gnuplot() returned 0: 0
gnuplot_show_png starting: 0
done read_plotspec_from_file: 0
get_png_bounds_info(): OK
calling render_pngfile: 0.015
done render_pngfile: 0.016
The second plot was faster, but even the first one took only about
0.8 seconds. The machine is a Lenovo T400 (Intel Core2 Duo CPU
P8600 @ 2.40GHz), fairly fast but nothing out of the ordinary.
If anyone else wants to try this, use tonight's Windows snapshot
and run gretlw32.exe with the --debug option. To get the full
effect of the timing code, open a time-series dataset and select a
single series for plotting. The timer output should appear in a
console window.
What I conclude from this is that there's no inherent problem in
the gretl/gnuplot/GTK code. If people are seeing very slow plot
times it seems that's an operating system issue.
Allin Cottrell