[pgrouting-dev] Support for Time Constraints
Stephen Woodbridge
woodbri at swoodbridge.com
Wed Jun 22 11:33:13 EDT 2011
Hi Jay,
For now just push forward with what you have for testing that should be
fine.
In general, the OSM data has lots of options for classification, but it
does not have clear standards or descriptions on how segments SHOULD be
classified, so it is easy for different people to interpret the
classifications differently. Hence, you end up with a huge list of
segment classes like on:
http://workshop.pgrouting.org/chapters/advanced.html
with only short 1-2 word descriptions. So for example what is the
difference between motorway, motorway_link, motorway_junction or for
that matter a lane, opposite, opposite_lane, track, byway, circleway,
living_street, residential, road, secondary, tertiary, track, roundabout
and grade* types and how should these be represented as classes or for
average speeds?
In the link above they are trying to deal with this same issue by
assigning a cost to the classes table based on the type name. They then
use that cost as a multiplier of the length which is weird, but kind of
works. This assumes the a cost multiplier of 2.0 means that that class
takes 2X as long to traverse that segment as a class with a cost
multiplier of 1.0.
UPDATE classes SET cost=2.0 WHERE name IN ('pedestrian','steps','footway');
versus:
UPDATE classes SET cost=0.3 WHERE name IN
('motorway','motorway_junction','motorway_link');
and in these segments vehicle speed is faster to the cost is smaller.
I take a more intuitive approach to this problem, by assigning an
average speed for each class of segment, then length / average speed
represents the average traversal time. So it is easy to make adjustments
to the model and you can easily do shortest time or distance.
-Steve
On 6/22/2011 10:42 AM, Jay Mahadeokar wrote:
>
>
> On Wed, Jun 22, 2011 at 7:57 PM, Stephen Woodbridge
> <woodbri at swoodbridge.com <mailto:woodbri at swoodbridge.com>> wrote:
>
> Hi Jay,
>
> I am back from my trip and wading through emails.
>
> I am interested in what your rationale for looking a length as a
> classifier is. It would seem to me the length of a given segment has
> nothing to do with classifying a segment with respect to speed or
> road class. The length of a segment is typically a function of how
> often that route is intersected by other features. For example and a
> path through a city will have more intersections than a path through
> a rural country side. A mountain pass could be a very long segment
> but might be a very slow path because it is twisty and has a lot of
> elevation changes. A major road through a city might have
> intersections at every block hence have short segments but could
> still move a lot of traffic from one side of the city to another.
>
> For testing purposes, any random classification is fine, but I like
> to pick classifications the reasonably model real life so when I see
> the results they reflect some general cognitive expectations and I
> can understand then or detect anomalies which might indicat a bug in
> the algorithm.
>
>
> Maybe I am missing the point, anyway just trying to understand and
> not suggesting that you change anything at this point.
>
>
> Hi Steve,
>
> Well I was very confused regarding the classification too since I have
> zero experience with real world data, and so I just choose length as
> criteria. This data is just for test purpose. We can modify the model
> and criteria any time, since writing the plsql procedure to generate
> such does not take too much effort. Also, the data does not affect the
> internal working of the algorithm, so I think for now, I can work with
> the currently generated test data.
>
> Any other ideas/ criteria for classification are welcome, and I would
> write the scripts to generate corresponding data.
>
>
> -Steve
>
>
> On 6/17/2011 12:42 AM, Jay Mahadeokar wrote:
>
> Hi,
>
> Here is an interesting look into the data in ways table:
>
> pgrouting-workshop=# select round(avg(length)*10000) as avg,
> round(min(length)*10000) as min, round(max(length)*10000) as
> max,count(*) as count, class_id from ways group by class_id
> order by avg
> desc;
>
> avg | min | max | count | class_id
> ------+------+-------+-------+ ----------
> 5320 | 250 | 12163 | 21 | 101
> 2135 | 1885 | 2385 | 2 | 111
> 1970 | 396 | 5168 | 17 | 102
> 1343 | 10 | 27485 | 234 | 106
> 1115 | 33 | 7964 | 221 | 108
> 1037 | 14 | 20020 | 248 | 109
> 938 | 9 | 23142 | 2761 | 110
> 837 | 21 | 2212 | 43 | 100
> 626 | 8 | 3327 | 90 | 112
> 619 | 26 | 9089 | 252 | 119
> 486 | 123 | 1618 | 11 | 117
> 473 | 7 | 16582 | 1259 | 114
> 364 | 97 | 1578 | 45 | 122
> 260 | 4 | 718 | 109 | 401
> (14 rows)
>
> So, I am going by the average length figure and considering
> class_id =
> (1001,111,102) as Type I (major_highways), which will be
> affected most
> by the traffic.
>
> class_id = (106,108,109,110,100) as Type II (minor_highways)
> which be
> comparatively less populated.
>
> class_id = (112,119,117,114,122,401) as Type III (streets) which
> will
> have least traffic.
>
> We want to have model such that during traffic hours the
> shortest path
> should go through streets.
>
> So, I am dividing day into following intervals, hope it sounds
> reasonable. I will write the corresponding plsql function that will
> generate the data accordingly. Note that the speed is in
> percentage of
> average speed. Since, we want to deal with time, we can effectively
> increase the lengths of the edges by same factor and assume speed as
> constant.
>
>
> Time Interval Speed for Type I Speed for Type II
> Speed for Type III
>
>
>
>
> 10 PM – 6 AM 100 100 100
> 6 AM – 7 AM 95 90 100
> 7 AM – 8 AM 80 75 90
> 8 AM – 9 AM 50 45 85
> 9 AM – 10 AM 60 55 85
> 10 AM – 11 AM 90 85 95
> 11 AM – 5 PM 100 100 100
> 5 PM – 6 PM 90 85 95
> 6 PM – 7 PM 75 70 90
> 7 PM – 8 PM 50 45 85
> 8 PM – 9 PM 60 55 85
> 9 PM – 10 PM
> 95 90 95
>
>
>
> Any feedback is welcome.
>
> I am currently stuck with a bug in the core time-dependent function,
> there seems to be a problem with non-integer data. Will try and
> fix that
> soon.
>
>
> On Wed, Jun 15, 2011 at 7:26 AM, Jay Mahadeokar
> <jai.mahadeokar at gmail.com <mailto:jai.mahadeokar at gmail.com>
> <mailto:jai.mahadeokar at gmail. com
> <mailto:jai.mahadeokar at gmail.com>>> wrote:
>
>
>
> On Sun, Jun 12, 2011 at 8:00 PM, Stephen Woodbridge
> <woodbri at swoodbridge.com <mailto:woodbri at swoodbridge.com>
> <mailto:woodbri at swoodbridge. com
> <mailto:woodbri at swoodbridge.com>>> wrote:
>
> Jay,
>
> Try:
>
> select class_id, count(*) as cnt from ways group by class_id
> oder by classid;
>
> Typically you should a a distribution in real world data
> something like:
>
> 5% limited access major highways
> 15% major roads
> 75% minor roads
> 5% trails, pedestrian ways, etc
>
>
> Hi,
>
> We have got following data in the pgrouting-workshop ways table.
>
> pgrouting-workshop=# select distinct(count(class_id)) as
> no_of_entries,class_id from ways group by class_id order by
> no_of_entries;
> no_of_entries | class_id
> ---------------+----------
> 2 | 111
> 11 | 117
> 17 | 102
> 21 | 101
> 43 | 100
> 45 | 122
> 90 | 112
> 109 | 401
> 221 | 108
> 234 | 106
> 248 | 109
> 252 | 119
> 1259 | 114
> 2761 | 110
> (14 rows)
>
>
> The classes table in routing database gives more information
> on the
> class_id field. Also available here:
> http://workshop.pgrouting.org/ chapters/advanced.html
> <http://workshop.pgrouting.org/chapters/advanced.html>
>
> So, we have:
>
> class_id name no_of_entries
>
> 110 track 2761
> 114 path 1259
> 119 steps 252
> 109 service 248
> 106 primary 234
> 108 living_street 221
> 401 roundabout 109
>
> These are the major class_types with more than 100 entries
> in ways
> table.
>
> Steve had earlier suggested following idea for time
> dependent data
> generation:
>
>
> "If we think about "rush hour" around a major city, the highways
> (based on road class) flowing into the city in the morning
> would get
> reduced average speeds you could apply curve like average
> speed*percent based on 6am (90%), 7am(75%), 8am(45%), 9am(50%),
> 10am(85%) and do something similar in the evening rush. It
> might be
> too hard to figure on direction of flow in/out bound so
> apply the
> curve to all traffic. The assumption is that the highways are
> congested which will force traffic onto side streets. You
> might want
> to also reduce the lower class speeds by say a constant 80%
> during
> rush hour."
>
> So, what classes should I consider as highways, what classes as
> streets and so on? Or What should be the pattern of speed
> changes in
> various classes?
>
> The main focus would be, we should be routed through less
> congested
> areas during rush hours right?
>
> Once we finalise this model, I can write the plsql function to
> generate the corresponding data.
>
>
>
> Well in the US anyway. Also if this is OSM data then you
> might
> get a better clue on classes by looking to that ....
> OH!, look
> what I found in google :)
>
> http://workshop.pgrouting.org/ chapters/advanced.html
> <http://workshop.pgrouting.org/chapters/advanced.html>
>
> Looks like first digit is type and the three digits are
> from the
> classes table.
>
> -Steve
>
>
> On 6/12/2011 4:33 AM, Jay Mahadeokar wrote:
>
>
>
> On Sun, Jun 12, 2011 at 1:32 PM, Anton Patrushev
> <anton.patrushev at georepublic. de
> <mailto:anton.patrushev at georepublic.de>
> <mailto:anton.patrushev@ georepublic.de
> <mailto:anton.patrushev at georepublic.de>>
> <mailto:anton.patrushev@ georepublic.de
> <mailto:anton.patrushev at georepublic.de>
> <mailto:anton.patrushev@ georepublic.de
> <mailto:anton.patrushev at georepublic.de>>>>
>
> wrote:
>
> Hi Jay,
>
> No, there was no specific meaning, just king of
> cyphering - first
> digit for road class and then two last digits
> for road type.
>
>
> Hi Anton,
>
> So, as you said last 2 digits are for road type. I
> can see
> they are
> mainly 00, 01, 02, 06, 08, 09, 10, 11, 12 ,14, 17.
>
> Road class is just 1 or 4.
>
> Can I differentiate major highways, minor highways,
> streets
> etc using
> this info?
>
> Anton.
>
> On 6/12/11, Jay Mahadeokar
> <jai.mahadeokar at gmail.com <mailto:jai.mahadeokar at gmail.com>
> <mailto:jai.mahadeokar at gmail. com <mailto:jai.mahadeokar at gmail.com>>
> <mailto:jai.mahadeokar at gmail. com <mailto:jai.mahadeokar at gmail.com>
> <mailto:jai.mahadeokar at gmail. com
> <mailto:jai.mahadeokar at gmail.com>>>> wrote:
> >> Hi Jay,
> >>
> >> I think that instead of just random times, I would take
> a different
> >> approach to generate this data. If we think about "rush
> hour"
> around a
> >> major
> >> city, the highways (based on road class) flowing into
> the city
> in the
> >> morning would get reduced average speeds you could apply
> curve like
> >> average
> >> speed*percent based on 6am (90%), 7am(75%), 8am(45%),
> 9am(50%),
> 10am(85%)
> >> and do something similar in the evening rush. It might
> be too
> hard to
> >> figure
> >> on direction of flow in/out bound so apply the curve to all
> traffic. The
> >> assumption is that the highways are congested which will
> force
> traffic
> >> onto
> >> side streets. You might want to also reduce the lower class
> speeds by say
> >> a
> >> constant 80% during rush hour.
> >>
> >> If we can get OSM data then it should be easy to
> populate the
> table with
> >> that data.
> >>
> >> -Steve
> >>
> >
> > I am trying to write plpgsql query to generate
> time-dependent data
> > corresponding to the ways table in pgrouting workshop. As
> suggested by Steve
> > above, instead of generating random data, I will follow
> patterns
> (see above
> > message) so that the data is nearer to the real worlds
> scenario.
> >
> > So, now I need to make distinction between highways,
> streets etc.
> I saw the
> > attribute class_id in ways table. It has 14 distinct values:
> > class_id
> >
> > ----------
> >
> >
> > 102
> >
> >
> > 122
> >
> >
> > 106
> >
> >
> > 111
> >
> >
> > 108
> >
> >
> > 100
> >
> >
> > 109
> >
> >
> > 112
> >
> >
> > 101
> >
> >
> > 110
> >
> >
> > 401
> >
> >
> > 119
> >
> >
> > 117
> >
> > 114
> >
> > Any specific meaning attached to these values? I did not
> find any
> > information on the pgRouting-workshop website [1].
> >
> >
> > Thanks in advance.
> >
> > [1] http://workshop.pgrouting.org/ chapters/topology.html
> <http://workshop.pgrouting.org/chapters/topology.html>
> >
> > --
> > Regards,
> > -Jay Mahadeokar
> >
>
>
> --
> Georepublic UG (haftungsbeschränkt)
> Salzmannstraße 44,
> 81739 München, Germany
>
> Anton Patrushev
> CTO
>
> eMail: anton.patrushev at georepublic.de
> <mailto:anton.patrushev at georepublic.de>
> <mailto:anton.patrushev@ georepublic.de
> <mailto:anton.patrushev at georepublic.de>>
> <mailto:anton.patrushev@ georepublic.de
> <mailto:anton.patrushev at georepublic.de>
> <mailto:anton.patrushev@ georepublic.de
> <mailto:anton.patrushev at georepublic.de>>>
>
> Web: http://georepublic.de
>
> Tel: +49 (089) 420 959 519
> Sip: 1959519 at sipgate.de
> <mailto:1959519 at sipgate.de> <mailto:1959519 at sipgate.de
> <mailto:1959519 at sipgate.de>>
> <mailto:1959519 at sipgate.de <mailto:1959519 at sipgate.de>
> <mailto:1959519 at sipgate.de <mailto:1959519 at sipgate.de>>>
>
>
>
> Commercial register: Amtsgericht München, HRB 181428
> CEO: Daniel Kastl
> ______________________________ _________________
> pgrouting-dev mailing list
> pgrouting-dev at lists.osgeo.org <mailto:pgrouting-dev at lists.osgeo.org>
> <mailto:pgrouting-dev at lists. osgeo.org
> <mailto:pgrouting-dev at lists.osgeo.org>>
> <mailto:pgrouting-dev at lists. osgeo.org
> <mailto:pgrouting-dev at lists.osgeo.org>
> <mailto:pgrouting-dev at lists. osgeo.org
> <mailto:pgrouting-dev at lists.osgeo.org>>>
>
> http://lists.osgeo.org/ mailman/listinfo/pgrouting-dev
> <http://lists.osgeo.org/mailman/listinfo/pgrouting-dev>
>
>
>
>
> --
> Regards,
> -Jay Mahadeokar
>
>
>
> ______________________________ _________________
> pgrouting-dev mailing list
> pgrouting-dev at lists.osgeo.org <mailto:pgrouting-dev at lists.osgeo.org>
> <mailto:pgrouting-dev at lists. osgeo.org
> <mailto:pgrouting-dev at lists.osgeo.org>>
> http://lists.osgeo.org/ mailman/listinfo/pgrouting-dev
> <http://lists.osgeo.org/mailman/listinfo/pgrouting-dev>
>
>
> ______________________________ _________________
> pgrouting-dev mailing list
> pgrouting-dev at lists.osgeo.org
> <mailto:pgrouting-dev at lists.osgeo.org>
> <mailto:pgrouting-dev at lists. osgeo.org
> <mailto:pgrouting-dev at lists.osgeo.org>>
> http://lists.osgeo.org/ mailman/listinfo/pgrouting-dev
> <http://lists.osgeo.org/mailman/listinfo/pgrouting-dev>
>
>
>
>
> --
> Regards,
> -Jay Mahadeokar
>
>
>
>
> --
> Regards,
> -Jay Mahadeokar
>
>
>
> ______________________________ _________________
> pgrouting-dev mailing list
> pgrouting-dev at lists.osgeo.org <mailto:pgrouting-dev at lists.osgeo.org>
> http://lists.osgeo.org/ mailman/listinfo/pgrouting-dev
> <http://lists.osgeo.org/mailman/listinfo/pgrouting-dev>
>
>
> ______________________________ _________________
> pgrouting-dev mailing list
> pgrouting-dev at lists.osgeo.org <mailto:pgrouting-dev at lists.osgeo.org>
> http://lists.osgeo.org/ mailman/listinfo/pgrouting-dev
> <http://lists.osgeo.org/mailman/listinfo/pgrouting-dev>
>
>
>
>
> --
> Regards,
> -Jay Mahadeokar
>
>
>
> _______________________________________________
> pgrouting-dev mailing list
> pgrouting-dev at lists.osgeo.org
> http://lists.osgeo.org/mailman/listinfo/pgrouting-dev
More information about the pgrouting-dev
mailing list