Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.graphics.apps.gnuplot > #1616 > unrolled thread

splot with dgrid3d / pm3d?

Back to article view | Back to comp.graphics.apps.gnuplot

Contents

  splot with dgrid3d / pm3d? neurino <lelli.luca@googlemail.com> - 2013-02-10 04:13 +0100
    Re: splot with dgrid3d / pm3d? sfeam <sfeam@users.sourceforge.net> - 2013-02-09 20:28 -0800
      Re: splot with dgrid3d / pm3d? neurino <lelli.luca@googlemail.com> - 2013-02-10 15:52 +0100
    Re: splot with dgrid3d / pm3d? Hans-Bernhard Bröker <HBBroeker@t-online.de> - 2013-02-10 16:49 +0100

#1616 — splot with dgrid3d / pm3d?

Dear gnuplotters,

i have 3D data, which look like this
# x y z
1996-12 68 -0.09928498
1996-12 69 -0.1597349
1996-12 70 -0.009438827
...
...
1997-01 68 -0.09928498
1997-01 69 -0.1597349
1997-01 70 -0.009438827

and so on, 192 data (x) blocks, each of 141 values (y).

The following plot
http://img39.imageshack.us/img39/8113/cthanzonalfull.jpg

has been created with (relevant commands):

set pm3d map
set isosample 1200
set xdata time
set timefmt "%Y-%m-%d"
splot f1 u 1:2:3 notitle

However, i'd like to smooth them and highlight patterns.
I am having difficult times with set dgrid3d, because if i set

set dgrid3d 40,40,16

i get this result
http://img16.imageshack.us/img16/1641/testiiq.jpg

So, from the demos i don't get how the non-gridded data should be 
preprocessed, having set dgrid3d instead of pm3d, given that the nice 
demos are achieved with splot 1:2:3 too.

Thanks for the help.

[toc] | [next] | [standalone]

#1617

neurino wrote:

> Dear gnuplotters,
> 
> i have 3D data, which look like this
> # x y z
> 1996-12 68 -0.09928498
> 1996-12 69 -0.1597349
> 1996-12 70 -0.009438827
> ...
> ...
> 1997-01 68 -0.09928498
> 1997-01 69 -0.1597349
> 1997-01 70 -0.009438827
> 
> and so on, 192 data (x) blocks, each of 141 values (y).

I may not understand the question.
It looks from the *.jpg you point to that you aren't really
trying for a 3D plot, but a 2D heatmap.  Is that correct?
If so then neither pm3d nor isosamples are needed.

Try  
    plot "data" using 2:1:3 with image

There are actually quite a few ways to create a heatmap.
You might have a look at the demos here

    http://gnuplot.sourceforge.net/demo_cvs/heatmaps.html

As to smoothing and highlighting, I'd have to know a lot
more about the data structure and what you are looking
for.  Maybe a Fourier transform + high frequency filter?
Or just output the heatmap as a PNG image and read it into
an image processing tool like Photoshop or GIMP.


	Ethan

> 
> The following plot
> http://img39.imageshack.us/img39/8113/cthanzonalfull.jpg
> 
> has been created with (relevant commands):
> 
> set pm3d map
> set isosample 1200
> set xdata time
> set timefmt "%Y-%m-%d"
> splot f1 u 1:2:3 notitle
> 
> However, i'd like to smooth them and highlight patterns.
> I am having difficult times with set dgrid3d, because if i set
> 
> set dgrid3d 40,40,16
> 
> i get this result
> http://img16.imageshack.us/img16/1641/testiiq.jpg
> 
> So, from the demos i don't get how the non-gridded data should be
> preprocessed, having set dgrid3d instead of pm3d, given that the nice
> demos are achieved with splot 1:2:3 too.
> 
> Thanks for the help.

[toc] | [prev] | [next] | [standalone]

#1618

On 2013-02-10 05:28:10 +0100, sfeam said:

Hello Ethan, thanks for the prompt reply.

> I may not understand the question.
> It looks from the *.jpg you point to that you aren't really
> trying for a 3D plot, but a 2D heatmap.  Is that correct?
> If so then neither pm3d nor isosamples are needed.
> 
> Try
>     plot "data" using 2:1:3 with image

line 0: warning: Number of pixels cannot be factored into integers 
matching grid. N = 52555  K = 50

> There are actually quite a few ways to create a heatmap.
> You might have a look at the demos here
> 
>     http://gnuplot.sourceforge.net/demo_cvs/heatmaps.html

Indeed. What i always thought was that a 3D plot, in map view 
perspective, is the analogue of a 2D heatmap. I realize only now that 
this might be wrong, though. With the following:

set pm3d implicit at s
set contour base
set cntrparam bspline
set cntrparam order 4
min = -0.5; max = 0.5
step=(max-min)/4
set cntrparam level discrete min,step,max
splot "data" using 1:2:3  w pm3d notitle

The result is here:
http://img825.imageshack.us/img825/5368/testpm3dimplicit.jpg

> As to smoothing and highlighting, I'd have to know a lot
> more about the data structure and what you are looking
> for.  Maybe a Fourier transform + high frequency filter?

Physically, the data are timeseries, sampled on a monthly basis (192 
months, the data blocks, the x-axis) and gridded from north to south at 
0.5 deg resolution (281 y-values).

So, while a FFT + Filter might be a sound option, but i wonder whether 
is overkill. It's just for visualization purpose. I achieve scientific 
insight with PCA (empirical orthogonal functions, covariance matrices 
and so on) anyway.

Given the last .jpg above, what can you recommend me? I never used FFT 
with gnuplot. Is it achievable?
Thanks.

[toc] | [prev] | [next] | [standalone]

#1619

On 10.02.2013 04:13, neurino wrote:

> has been created with (relevant commands):
>
> set pm3d map
> set isosample 1200

This is quite irrelevant for a pure data plot.

> So, from the demos i don't get how the non-gridded data should be
> preprocessed,

But your data _is_ gridded!  Or at least, it very easily could be.   All 
it would take is for you to indicate where in your data each new block 
starts, by inserting a blank line.

 > having set dgrid3d instead of pm3d,

Why "indstead of"?  Those are totally separate options.

[toc] | [prev] | [standalone]

Back to top | Article view | comp.graphics.apps.gnuplot

csiph-web