[pgrouting-users] Setting the tolerance argument in assign_vertex_id

[Stephen Woodbridge]

Dan,

I think that picking a number around 0.5 to 1 meter should be fine for 
that data. In general it is unlikely that you will have nodes that are 
NOT connected but that close together.

You can also do some analysis of the results like this, after you run 
assign_vertex_id():

alter table vertices_tmp add column cnt int;
update vertices_tmp set cnt = (select count(*) from "<edge_table>"
    where vertices_tmp.id=target or vertices_tmp.id=source);
select cnt as connections, count(*) from vertices_tmp
    group by cnt order by cnt;

This will analyze the connectedness of you network.

connections
   1   - dead ends
   2   - segments connected but only an intersection if different names
   3+  - intersections

If you run assign_vertex_id() with different tolerance values and then 
run this analysis as your tolerance gets too big you will see a shift of 
these numbers to the smaller end which is bad.

Another important analysis you check is:

select count(*) from "<edge_table>" where target=source;

this count should be zero unless you have zero length segments in your 
data and you can check that with:

select gid, st_length(the_geom) from "<edge_table>" where target=source;

These are a good way to learn about your data.

-Steve


On 8/26/2010 8:37 PM, Dan Putler wrote:
> Thanks Steve, but I do need a bit of a follow on I think, and
> probably provide a bit more explanation.
>
> I'm using a DMTI CanMap routing layer for BC that started out in
> NAD83 geographic, which I converted to NAD83 UTM Zone 10N via
> ogr2ogr, I also shrank it down using a "with" statement in ogr2ogr to
> cover only part of the province. After I sent my email, I did a bit
> more Google searching and based on a question in the pgRouting forum,
> a user using NAVTEQ layers was using a tolerance that worked out to
> between 1 and 1.5 meters. Based on this, I set the tolerance to 2
> meters, it sounds like I should set the tolerance in assign_vertex_id
> to be much smaller (say 0.15 meters, which is just over 5 inches).
>
> In general, is it better to err on the side of too small or too large
> a tolerance, or is that a "it depends" question?
>
> Dan
>
> --- On Thu, 8/26/10, Stephen Woodbridge<woodbri at swoodbridge.com>
> wrote:
>
>> From: Stephen Woodbridge<woodbri at swoodbridge.com> Subject: Re:
>> [pgrouting-users] Setting the tolerance argument in
>> assign_vertex_id To: "Dan Putler"<putler at yahoo.com> Cc: "pgRouting
>> list"<pgrouting-users at lists.osgeo.org> Received: Thursday, August
>> 26, 2010, 4:27 PM On 8/26/2010 2:51 PM, Dan Putler wrote:
>>> Hi,
>>>
>>> I'm curious how to determine the value of the second
>> argument in the
>>> function assign_vertex_id(). Most of the examples I
>> can find look
>>> like:
>>>
>>> assign_vertex_id('<edge table>', 0.001,
>> 'the_geom', 'gid')
>>>
>>> Most of the examples are based on edge tables that
>> have SRIDs that
>>> correspond to either WGS84 geographic or some sort of
>> US State Plane
>>> feet. I'm using data that is using a UTM coordinate
>> SRID, so the map
>>> units are in meters. My question is whether the map
>> units should
>>> influence the choice of the second argument to the
>> function (which I
>>> assume is a tolerance between edges), or should I just
>> keep with the
>>> value 0.001? Implicitly, I'm asking if the parameter
>> is in the map
>>> units of the SRID.
>>
>> Dan,
>>
>> Right units is important.
>>
>> When I use data in WGS84 that has at lease 6 decimal places then I
>> use 0.000001 as the tolerance. What this means if that if the
>> distance between two points are less than or equal 0.000001 units
>> then they should be considered the same point. 0.000001 degrees ==
>> ~5 inches if I did the math right.
>>
>> I think there are two issues you need to be aware of:
>>
>> 1) units 2) your data and what precision it was built at
>>
>> HTH, -Steve
>>
>>
>
>