matplotlib-users Mailing List for matplotlib

Nicolas,
Please check the postscript backend results regardless of whether the 
cntr.c change I suggested makes a difference or not; in fact, just try 
the different backend first. You could also try the png backend with a 
different dpi setting. Looking again at your plot, it looks a bit like a 
rounding problem, and I don't see how that would arise in cntr.c. It 
could still be a rendering problem and be related to the chunking; there 
might be rendering anomalies at the chunk boundaries. Cntr.c itself is 
still the prime candidate, though. The artifacts remind me somewhat of 
the bug I was working on; maybe in both cases, the tracing algorithm is 
turning left when it should turn right, or something like that. It 
might be something like a "<" that should be a "<="...
Near the very bottom of the Bz and some other panels there are some 
spiky features, different from the blocky artifacts that make stripes 
near 0.3, etc. Are the little spiky things in the data, or are they 
also artifacts?
Do the blocky artifacts correspond to your grid spacing--are they 
single-gridpoint offsets?
If you plot only a subdomain--a smaller range of x and y--do the 
artifacts remain?
Eric
> 
> Now, my first question is: notice the artifacts on the contours of Bz and I at 
> y=0.3 and y=0.55, and on the contours of Q for multiples values of y. These 
> doesn't come from data, as they disappear when zooming in. Do you know where 
> do they come from ? And, more importantly (to me), how to get rid of them ?

Hello John & NG,
 sorry about the code wrapping, I don't know what 
happens to my email... I have to attach it in a Python file in this 
email, because my WebMail always wraps badly the code.
Thank you for a 
pointer to the Wiki!
Andrea.
"Imagination Is The Only Weapon In The War Against Reality."
http://xoomer.virgilio.it/infinity77

Nicolas,
I don't remember seeing artifacts like those before. Most likely they 
come from a bug in the basic contouring routine, cntr.c, which is 
derived with slight modifications from the routine gcntr.c in gist. 
Beyond that, its origins seem lost in the murky past. It seems to be 
old, clever, complicated, and very hard to understand in detail. A few 
months ago I spent quite a bit of time trying to track down another bug, 
and I failed.
It might be related to chunking. In cntr.c, try changing line 1384
 long nchunk = 30; /* hardwired for now */
to a different value--say 50--and see if the artifacts change. If they 
don't, verify that the artifacts are the same when the postscript 
backend is used.
Good free contouring routines, especially ones that do filled contours 
and handle data masks, seem remarkably hard to find.
Eric
> I'd like to ask you several questions related to a figure I've generated ; I 
> apologize in advance for the forthcoming messages, because I've chosen to 
> start a new thread for each question...
> 
> My figure is located at http://nicolasgirard.nerim.net/img/artifacts.png . 
> Beware that the file is quite large (over 400 kb).
> 
> Now, my first question is: notice the artifacts on the contours of Bz and I at 
> y=0.3 and y=0.55, and on the contours of Q for multiples values of y. These 
> doesn't come from data, as they disappear when zooming in. Do you know where 
> do they come from ? And, more importantly (to me), how to get rid of them ?
> 
> Cheers,
> Nicolas