Re: Undocumented behaviour of "smooth" options

[<pl...@pi...>]

On 21/02/17 23:21, Ethan A Merritt wrote:
>
> On the bug tracker    <https://sourceforge.net/p/gnuplot/bugs/1911/>
> Rainald Koch points out that when a data set contains some points with
> invalid y values, the result of smoothing is unexpected. This is true
> also if you use a filter to select a subset of the points, e.g.
>
>     plot FOO using 1:( filter($2) ? $2 : 1/0) smooth acsplines
>
> I can't find any mention of it in the documentation, but it seems that
> this is intentional.  The "smooth" code has a pre-processing step that
> splits the input data into N subsets separated by one or more "undefined"
> points. The smoothing is then applied separately to each of the N segments
> with no attempt to make the fit smooth or continuous at the junctions
> between segments.

I can see that this will be a problem with some splines which tend to be 
unstable at the end points and can veer off in either direction in a 
very unsmooth fashion.

In passing, I would even question whether linking to the end points of 
Bezier spline is even desirable. This seems poorly constrained and gives 
unsatisfactory results. Test with a limited dataset of say 20 points to 
get the idea.

This raises the question of why the code is splitting otherwise 
continuous data. Is this required to reduce computation for some of the 
smoothing methods, is it required for all types of smoothing algo. ?

If  this does  give unexpected results for datasets with breaks , as 
reported, is this a bug or a feature?  Would a smoothing process be 
expected to smooth and link across breaks in the data. Perhaps the 
pre-processing should be removing NaNs non inserting them.

Should this be an option to "smooth" ?

Peter.



>
> Is this documented somewhere that I haven't found?
>
> Is there a recommended way to fit a single smooth curve to the
> entire set of points after filtering?
>
> Current CVS allows this work-around:
>
>    set datafile missing NaN
>
> but that is very recent.  Was there no way to handle this in the past?
>
> 	Ethan
>
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, SlashDot.org! http://sdm.link/slashdot
> _______________________________________________
> gnuplot-beta mailing list
> gnu...@li...
> Membership management via: https://lists.sourceforge.net/lists/listinfo/gnuplot-beta
>