[yt-dev] A sketch for a cut_region selector

[Nathan Goldbaum]

On Tue, Nov 18, 2014 at 10:58 PM, Nathan Goldbaum <nathan12343 at gmail.com>
wrote:

> Hi all,
>
> I spent some time tonight thinking of a way to supply a cut_region as a
> data_source for a SlicePlot.
>
> Right now this will raise a NotImplementedError, since cut_regions don't
> work using selectors. Instead, they override data selection at the level of
> the chunking system.  This is a problem for a SlicePlot, since a slice is
> really a 2D selection object and makes use of a selector object to decide
> whether or not data should be included in the slice. Cut regions
> technically could be implemented using selectors, but they don't lend
> themselves to the geometrically focused selector API.
>
> One way around this is to eat some memory and pass the positions of cells
> that pass the cut_region selection criteria down to the cython selector. Lo
> and behold, it actually works!   Kind of.
>
> I've pasted the necessary diff here:
> http://paste.yt-project.org/show/5210/
>
> Here's an example script: http://paste.yt-project.org/show/5211/
>
> After much gnashing of teeth (about 20 minutes worth on my laptop), this
> should produce the following image: http://imgur.com/DlHruOi
>
> The way I've done this is really bad: not only am I re-aquiring the GIL
> and messing with python objects inside a tight cython function, but I'm
> also using the python objects to do an O(N^2) search algorithm!
>
> I think it's possible to speed this up significantly by using a smarter
> search algorithm implemented in C.  Before I spend effort on that, I wanted
> to send this e-mail out to summarize my thoughts and check to see if this
> is a crazy approach.
>
> If this is a worthy approach, does anyone know of an O(logN) search
> algorithm I could use to find a 3D position in a list of 3D positions?
>

Woah, it's *a lot* faster if I use a python dict as a hashmap.  It seems
checking if a key is in a dict is O(1), and that helps a lot.  I think I'm
going to PR this!


>
> Thanks for taking a look :)
>
> -Nathan
>
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.spacepope.org/pipermail/yt-dev-spacepope.org/attachments/20141119/874babd1/attachment.html>