[ncl-talk] issue with gsn_add_polygon

[Dave Allured - NOAA Affiliate]

Okay.  That will work, but it will draw the first bin asymmetric with
respect to the rest of the polygon, except when pdfib(0) = 0.  For very
small pdfib(0), the asymmetry will not be noticeable.  For a true
horizontal edge at the base of each polygon, you should include the extra
coordinate pair like I showed before.


On Sun, Jun 14, 2020 at 2:20 PM Adam Herrington <
adam.herrington at stonybrook.edu> wrote:

> You're right, they are not closed contours. I forced them closed through
>
>     xib = new( (/2*nbins/), "double")
>     yib = new( (/2*nbins/), "double")
>     ...
>     xib(0) = xib(2*nbins-1)
>     yib(0) = yib(2*nbins-1)
>
> Thanks a lot Dave! I'll be presenting this plot at cesm workshop on
> wednesday if you're curious of what it means.
>
> Adam
>
> On Sun, Jun 14, 2020 at 8:18 AM Dave Allured - NOAA Affiliate <
> dave.allured at noaa.gov> wrote:
>
>> Very nice plots.  Notice how those mystery lines fade as they approach
>> (0,0).  That is because they are polygons, not lines.  Specifically, long
>> skinny triangles.  That is a clue.
>>
>> xib and yib are not closed contours.  I am not sure whether
>> gsn_add_polygon is supposed to close them for you.  But since this seems to
>> coincide with the visible anomalies, try closing them explicitly.
>>
>>     xib = new( (/2*nbins+1/), "double")
>>     yib = new( (/2*nbins+1/), "double")
>>     ...
>>     xib(2*nbins) = xib(0)
>>     yib(2*nbins) = yib(0)
>>
>>
>> On Fri, Jun 12, 2020 at 7:59 PM Adam Herrington via ncl-talk <
>> ncl-talk at mailman.ucar.edu> wrote:
>>
>>> I'm attempting to compute a raw violin-type plot. This is similar to a
>>> box plot, but I'm drawing a polygon around the pdf, and I close the polygon
>>> through a mirror image (-1*probability). Then I rotate it so it's vertical,
>>> and align it with the x-axis "id" of the box plot.
>>>
>>> If you're not following, I don't blame you ... this is hard to describe.
>>> Attached is my attempt at this that will hopefully clarify what I'm trying
>>> to do.
>>>
>>> **digression before I go on to details on my primary issue**
>>> I'd like to experiment with an actual violin plot, with a smoothed
>>> kernel density estimate of the distribution. The routine kde_n_test
>>> <https://www.ncl.ucar.edu/Document/Functions/Built-in/kde_n_test.shtml> should
>>> work, but the warnings on the ncl page say the function is under
>>> construction. Is that the latest update for this function?
>>> **end digression**
>>>
>>> You'll notice for panels 2 and 3 there is a line pivoting from the
>>> bottom of the pdf's to the (0,0) point on the plot. I'm having trouble
>>> tracking down why this occurs and how to fix it. I'm going to copy-paste
>>> the relevant portions of the code for one of the x-value id's.
>>>
>>>   dx = 0.4d0 ;; x-depth of the pdf
>>>   xvals = (/1.,2.,3.,4.,5.,6./) ;; case id
>>> ...
>>>     pdfib = pdfx(ibdiff,nbins,optpdf)
>>>     pdfib = pdfib/max(pdfib)
>>>
>>>     xib = new( (/2*nbins/), "double")
>>>     yib = new( (/2*nbins/), "double")
>>>
>>>     xib(0:nbins-1) = xvals(n) - dx*pdfib
>>>     xib(nbins:2*nbins-1) = xvals(n) + dx*pdfib(nbins-1:0)
>>>     yib(0:nbins-1) = pdfib at bin_center
>>>     yib(nbins:2*nbins-1) = pdfib at bin_center(nbins-1:0)
>>>
>>>     pgstr = unique_string("dum")
>>>     plot@$pgstr$ = gsn_add_polygon(wks,plot(0),xib,yib,respg)
>>>
>>> I am pretty sure the xib and yib are a closed contour.. they go from
>>> 0:nbins-1 and then nbins:2*nbins-1 ... Would like to get another set of
>>> eyes on this though
>>>
>>> thanks,
>>> Adam
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.ucar.edu/pipermail/ncl-talk/attachments/20200614/d07e1db9/attachment.html>