[postgis-devel] More Cascade Union Adventures

Obe, Regina robe.dnd at cityofboston.gov
Wed Aug 13 16:03:35 PDT 2008


Hmm that doesn't look like the right dataset.  Now I have to figure out where I dug up this other towns layer.  Running on this particular layer ST_CascadeUnion runs in 
10 secs.  Regular ST_Union takes 99,703 ms. 

I'll let you know when I find it.  I have too many places where I have servers and all have slightly different datasets with the same name.



-----Original Message-----
From: postgis-devel-bounces at postgis.refractions.net on behalf of Obe, Regina
Sent: Wed 8/13/2008 6:40 PM
To: PostGIS Development Discussion
Subject: RE: [postgis-devel] More Cascade Union Adventures
 
Sure - its just the one on the MassGIS site

ftp://data.massgis.state.ma.us/pub/shape/state/towns.exe

Shucks - I guess that means I don't win a prize huh :(

Thanks,
Regina

-----Original Message-----
From: postgis-devel-bounces at postgis.refractions.net on behalf of Martin Davis
Sent: Wed 8/13/2008 6:40 PM
To: PostGIS Development Discussion
Subject: Re: [postgis-devel] More Cascade Union Adventures
 
I'm pretty sure that OJ is *not* calling CascadeUnion, but is using a 
special-purpose, similar algorithm developed by Michael Michaud.   I 
just looked at the OJ src code to confirm, and this looks to be the 
situation.  I think he tested his code against the JTS CascadedUnion and 
decided that the JTS algorithm is still faster. 

I just tried the OJ Union as well as the JTS CascadedUnion on the 
32278-feature dataset.  Results were:

OJ Union: 27 sec
JTS CascUnion: 17.7 sec

So JTS still wins, but they're not too far apart.

Just for grins, can you send me your Mass towns dataset to try?

Obe, Regina wrote:
>
> Martin,
>
> Just loaded that plugin - unless I put it in the wrong folder. - it 
> just seemed to create an option called aggregation-options under 
> plugins which seems to require two layers and does union, count 
> between the 2 layers.  So nope doesn't sound like Cascaded union and 
> didn't seem to change anything with the union function.
>
> If OpenJump union function isn't doing cascaded union, then what is 
> that count down thing for.  Doing the 30,000 it displays something 
> like this
>
> Computing Union
> 1/8 (32,278)
> 2/8 (8070)
> 3/8 ... /2018
> 86/505
> 1/127
> 6/8 (2/32)
> 8/8 (2/3)
> 3/3 (8/8)
>
> I also thought its speed of unioning of Mass towns was pretty impressive.
>
> Thanks,
> Regina
> -----Original Message-----
> From: postgis-devel-bounces at postgis.refractions.net on behalf of Obe, 
> Regina
> Sent: Wed 8/13/2008 5:40 PM
> To: PostGIS Development Discussion
> Subject: RE: [postgis-devel] More Cascade Union Adventures
>
> I assumed it was since it was counting down like in some sort of 
> upside down pyramid
>
> 500
> 255
> 10
> :
> :
>
> So it seemed like it was doing some sort of division of the 
> geometries.  I'll give the below
> a try.  Maybe I misunderstood what that counting was for.
>
>
>
> -----Original Message-----
> From: postgis-devel-bounces at postgis.refractions.net on behalf of 
> Martin Davis
> Sent: Wed 8/13/2008 4:49 PM
> To: PostGIS Development Discussion
> Subject: Re: [postgis-devel] More Cascade Union Adventures
>
> Yep, I don't Hausdorff distance is available in very many places.  Pity,
> because it is a very useful metric for comparing geometry.  The full
> Hausdorff distance is quite challenging to implement, but I have made a
> VertexHausdorffDistance approximation which is just as useful in most
> situations, and is a lot simpler to implement (and faster to run).
>
> By the way, are you sure that OpenJUMP is using CascadedUnion?  AFAIK it
> didn't in the past... Michael Michaud has just released (today!) an
> extension which I think does use CascadedUnion - so you might want to
> try that.
>
> http://geo.michaelm.free.fr/OpenJUMP/resources/aggregation-0.1.jar
>
> When I did the original testing with the 30K polygon dataset that you're
> using, I was getting times of around 20 sec using CascadedUnion...
>
>
>
> Obe, Regina wrote:
> >
> > Martin,
> > Pardon my ignorance.
> >
> > I don't see a Hausdorff distance in OpenJump or maybe it goes by a
> > more verbose name and the descriptions I read about Hausdorff
> > distances are greek to me.
> >
> > Well the Mass town test seems to pass my trivial test exercises. It
> > looks like massachusetts with no visually apparent gaps, has the same
> > number points in all cases, similar area
> >
> > both num points - 476026
> > both num geometries - 694
> > area (ST_CascadeUnion - 2.094208570266725E10)
> > area (JTS - 2.0942085702666965E10)
> >
> >
> > -----Original Message-----
> > From: postgis-devel-bounces at postgis.refractions.net on behalf of
> > Martin Davis
> > Sent: Wed 8/13/2008 11:51 AM
> > To: PostGIS Development Discussion
> > Subject: Re: [postgis-devel] More Cascade Union Adventures
> >
> > I wouldn't expect to have the results be exactly equal - the union code
> > is likely to be input-order dependent.
> >
> > And the difference in areas doesn't seem surprising either - it's way
> > down in the small decimal places, which would occur even with slight
> > differences in the geometry.
> >
> > A more revealing test would be to compute the Hausdorff distance between
> > the union boundaries - that would show if they differed by very much,
> > and where.  PostGIS doesn't have this - I can't remember whether
> > OpenJUMP does or not.  JEQL has this operation, too.
> >
> > Obe, Regina wrote:
> > > Now testing all including JTS 1.9.0 (OpenJump) on a Win XP runing
> > > PostgreSQL 8.3.1, PostGIS 1.3.3, Geos 3.0.0.  It appears using array
> > > trumps all, Cascade aggregate union is a vast improvement over 
> ST_Union
> > > (ST_Union I didn't bother testing of course because it owuld never
> > > finish on this test), but evidentally the array accum calls give a 
> major
> > > penalty.
> > >
> > > I found some things I found a bit possibly disturbing.  Maybe its just
> > > the nature of unioning in different orders.  I compared all 3 outputs
> > > and none of them are binary equal or even ST_Equals for that matter.
> > >
> > > However all 3 give same basic stats using OpenJump - e.g
> > > All result in = 32972 pts   
> > > components = 10
> > > lengths and areas are off by a bit
> > > length = agg cascade union = 17.262684721407624, k nested union =
> > > 17.262684721407688,
> > >       jts = 17.262684721406348
> > >
> > > Should I be bothered by any of these or are they just rounding errors?
> > >
> > > --Other odd thing is that the array approach is not as good as it 
> was on
> > > my other machine bu the aggregate union performs better.  I'll just
> > > chuck this off to different postgresql version, and memory settings.
> > >
> > > Thanks,
> > > Regina
> > >
> > >
> > > --   209,563 | 198,391 ms - SELECT 198391/1000.00/60 = 3.31 minutes
> > >
> > > SELECT ST_CascadeUnion(the_geom)
> > > FROM (SELECT the_geom FROM sample_poly) As foo;
> > >
> > > -- 48,594 ms | 47,688 ms = 48 secs
> > > SELECT st_unitecascade_garray_sort(ARRAY(SELECT the_geom FROM
> > > sample_poly));
> > >
> > > -- 74,515 ms | 74,922 ms = SELECT 74515/1000.00/60 = 1.24 minutes
> > > SELECT ST_Union(the_geom) AS the_geom, 'nested union'
> > > FROM (
> > >    SELECT min(id) AS id, ST_Union(the_geom) AS the_geom
> > >    FROM (
> > >      SELECT min(id) AS id, ST_Union(the_geom) AS the_geom
> > >      FROM (
> > >        SELECT min(id) AS id, ST_Union(the_geom) AS the_geom
> > >        FROM (
> > >          SELECT min(id) AS id, ST_Union(the_geom) AS the_geom
> > >          FROM (SELECT the_geom, id FROM sample_poly) As foo
> > >          GROUP BY round(id/10)
> > >          ORDER BY id) AS tmp1
> > >        GROUP BY round(id/100)
> > >        ORDER BY id) AS tmp2
> > >      GROUP BY round(id/1000)
> > >      ORDER BY id) AS tmp3
> > >    GROUP BY round(id/10000)
> > >    ORDER BY id) AS tmp4
> > > GROUP BY round(id/100000);
> > >
> > >
> > > -- Open Jump JTS 1.9.0 Win XP
> > > --After loading running union across the whole set
> > > --1.14 minutes
> > > --Loading database query
> > > SELECT ST_AsBinary(the_geom)
> > > FROM sample_poly;
> > >
> > > -----------------------------------------
> > > The substance of this message, including any attachments, may be
> > > confidential, legally privileged and/or exempt from disclosure
> > > pursuant to Massachusetts law. It is intended
> > > solely for the addressee. If you received this in error, please
> > > contact the sender and delete the material from any computer.
> > >
> > > _______________________________________________
> > > postgis-devel mailing list
> > > postgis-devel at postgis.refractions.net
> > > http://postgis.refractions.net/mailman/listinfo/postgis-devel
> > >
> > > 
> >
> > --
> > Martin Davis
> > Senior Technical Architect
> > Refractions Research, Inc.
> > (250) 383-3022
> >
> > _______________________________________________
> > postgis-devel mailing list
> > postgis-devel at postgis.refractions.net
> > http://postgis.refractions.net/mailman/listinfo/postgis-devel
> >
> >
> >
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > postgis-devel mailing list
> > postgis-devel at postgis.refractions.net
> > http://postgis.refractions.net/mailman/listinfo/postgis-devel
> >  
>
> --
> Martin Davis
> Senior Technical Architect
> Refractions Research, Inc.
> (250) 383-3022
>
> _______________________________________________
> postgis-devel mailing list
> postgis-devel at postgis.refractions.net
> http://postgis.refractions.net/mailman/listinfo/postgis-devel
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> postgis-devel mailing list
> postgis-devel at postgis.refractions.net
> http://postgis.refractions.net/mailman/listinfo/postgis-devel
>   

-- 
Martin Davis
Senior Technical Architect
Refractions Research, Inc.
(250) 383-3022

_______________________________________________
postgis-devel mailing list
postgis-devel at postgis.refractions.net
http://postgis.refractions.net/mailman/listinfo/postgis-devel



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-devel/attachments/20080813/a643b7ea/attachment.html>


More information about the postgis-devel mailing list