Re: GSOC: Student intro, geometrical SP-GIST

Hi!

On Fri, Mar 20, 2015 at 5:00 PM, Dima Ivanovskiy <dima-iv(at)mail(dot)ru> wrote:

> Hello
> I am Dmitrii Ivanovskii, student of Moscow Institute of Physics and
> Technology
>
> I saw list ideas of gsoc and chose "Indexing prolonged geometrical objects
> (i.e. boxes, circles, polygons, not points) with SP-GiST by mapping to
> 4d-space".
>
> As I understand we already have GIST for geo data,
> http://postgis.net/docs/using_postgis_dbmanagement.html#gist_indexes
> But still don't have SP-GIST
>
> Question:
> 1. Who can be mentor of this project, because maybe I will have a lot of
> question about implementation B-Tree, R-tree in Postgresql, and I can talk
> with him.
> E-mail for starting will be good.
>

I can be possible mentor. You can talk to me.

2. I understand that Postgresql developers have a lot of different ideas
> and don't have a lot of time to do full describe of it. But I need all
> information about this idea. Of cource I try to search, but if author of
> the idea write more information it will be good
> 3. Will Index uses 4-d-space for all examples of objects? I still don't
> understand concept of this mapping. Circle in 2-D space have only 3
> coordinates(x, y, R)
> 4. When we think about "prolonged geometrical objects" is it only
> plane(2-D) object or more dimensional. I mean point->circle->sphere->...
> 5. I'm not sure whether that makes a difference. Do we talk about
> geometry with the Cartesian coordinate system? I saw that postgresql has
> modules for working
> with the geometry on the sphere.
>

This project idea is to research applicability of SP-GiST to non-point
objects.
The given sentence "Indexing prolonged geometrical objects (i.e. boxes,
circles, polygons, not points) with SP-GiST by mapping to 4d-space" is only
one of possible approaches for that task. In R-tree object is approximated
by MBR. MBR for 2d-objects can be mapped to 4d-point. More general,
nd-object MBR can be mapped into 2nd-point. Such points could be indexed by
SP-GiST. However, tree traversal for spatial queries would be non-trivial
and could be slow. So, this is only one of possible approaches.
I expect GSoC project on this subject to try several approached for this
task including one described above. But this is research project. It's
possible that none of approaches have majority over GiST.

------
With best regards,
Alexander Korotkov.