Skip to content

Latest commit

 

History

History
112 lines (73 loc) · 6.67 KB

README-OTP2.md

File metadata and controls

112 lines (73 loc) · 6.67 KB

ENTUR Point-to-point routing based on r5

Her is a temporary description of the ENTUR project to create a point-to-point routing engine. The paln is to integrate this into OTP, but to speed up development, the development is done here in the r5 GitRepo - not in OTP - for the time being.

Running the Speed Test

We use a test to verify the performance and test different travel searches. This is the best way to explore the algorithm. The SpeedTest gets it input from the travelSearch.csv file.

Test data setup

You need to have:

  1. A GFTS zip file
  2. A OSM map file for the same area.
  3. A CSV test data file with one routing request per line(travelSearch.csv).

The best way to get started is to run with the provided Norwegian test data set.

  1. Copy the [Git repo root]/testdata/norway directory to you prefered testdata directory, <test-data-dir>. Some of the needed files are checked in, but you must download GTFS transit data and a osm map.
  2. We did NOT provide the OSM map for norway so you have to download it. Download norway-latest.osm.pdf from
    geofabrik.de. Move it to <test-data-dir>.
  3. Take a look at the <test-data-dir>/travelSearch.csv, each line is a search and is run as part of the test.

That's it!

Building a Network file

Run: java -Xmx12G <cp> com.conveyal.r5.R5Main point --build <test-data-dir>

It is possible that you can build the graph with less memory than 12G.

This creates a network.dat file in your <test-data-dir> directory.

Running the SpeedTest

  1. Edit the <test-data-dir>/travelSearch.csv to match your test data
  2. Edit the SpeedTest#buildDefaultRequest(), set at least the date.
  3. Run the test.

Speed test program options

Run: java -Xmx12G <cp> com.conveyal.r5.speed_test.SpeedTest -h

This should output something like this:

usage: [options]
 -c,--testCases <arg>                   A coma separated list of test case numbers to run.
 -d,--dir <arg>                         The directory where network and input files are located. (Optional)
 -D,--debug                             Enable debug info.
 -h,--help                              Print all command line options, then exit. (Optional)
 -i,--numOfItineraries <arg>            Number of itineraries to return.
 -m,--searchTimeWindowInMinutes <arg>   The time in minutes to add to from to time.
 -n,--sampleTestNTimes <arg>            Repeat the test N times. Profiles are altered in a round robin fashion.
 -p,--profiles <arg>                    A coma separated list of configuration profiles:
                                        'original' or 'or' : The original R5 FastRaptorWorker by Conveyal
                                        'range_raptor' or 'rr' : Standard Range Raptor, super fast [ transfers, arrival time, travel time ].
                                        'mc_range_raptor' or 'mc' : Multi-Criteria pareto optimal Range Raptor [ transfers, arrival time,
                                        travel time, walking ].
 -s,--debugStops <arg>                  A coma separated list of stops to debug.
 -t,--debugTrip <arg>                   A coma separated list of stops representing a trip/path to debug. Use a '*' to indicate where to
                                        start debugging. For example '1,*2,3' will print event at stop 2 and 3, but not stop 1 for all trips
                                        starting with the given stop sequence.
 -v,--verbose                           Verbose output, print itineraries.

The arguments might be slightly different in your version, but the most useful arguments are:

  • -d <test-data-dir>
  • -p rr, -p mc, or -p mc,rr Run the test with Standard Range Raptor(fast), Pareto-optimal-multicriteria Range Raptor (slow), or both to compare results.
  • -v Print all paths/itineraries even if the test passes. Itineraries for failing tests are always printed.
  • -n 8 Repeat the test 8 times to get better performance measurement results (The -p rr is so fast that you need to repeat the test to measure the performance more accurate).
  • -c 21 or -c 3,7,21 Run a given set of tests. Use the testcaseId from the travelSearch.csv file.
  • -t 86128,86121,84932 Debug a trip. This print every event affecting the specified trip, this is great for understanding how raptor works and to find out why a trip is dropped in favor of another trip.

Setting the request parameters

Some of the request parameters are hard-coded in the speed test. Edit the SpeedTest#buildDefaultRequest() to match your transit data, at least the date must be a valid service date.

Run the speed test

To run the test with using the super fast Standard Range Raptor algorithm use:

Run: java -Xmx12G <cp> com.conveyal.r5.speed_test.SpeedTest -d <test-data-dir> -p rr

Comparison and verification of the result

The SpeedTest read the expected results from the travelSearch-expected-results.csv file. The expected results are compared with the actual result found in the travel search. The SpeedTest also write its result to a new file, the travelSearch-expected-results-out.csv each time it run.

If there is a mismatch between the expected and the actual the diff is printed to standard error:

SpeedTest FAILED     27 ms  #21 Indre Billefjord - Karasjok bussterminal, (70,315, 25,047) - (69,473, 25,517)  - Test assert errors  (TestCaseFailedException) 
      STATUS       | TF | Duration | Walk |   Start  |    End   | Modes | Agencies | Routes  | Stops             | Legs                                                                                      
WARN! Not expected |  1 |  1:56:12 |   12 | 09:08:59 | 11:05:11 |  BUS  |    Sne   | 066 063 | 86128 86121 84932 | WALK 0:01 - 86128 - BUS 066 09:10 09:45 - 86121 - BUS 063 10:00 11:05 - 84932 - WALK 0:11 
 FAILED! Expected  |  1 |  2:10:33 | 1051 | 08:54:38 | 11:05:11 |   -   |     -    | -       | -                 | WALK 13:22 - 86061 - BUS 066 09:09 09:45 - 86121 - BUS 063 10:00 11:05 - 84932 - WALK 0:11
        OK         |  0 |  2:10:33 | 1051 | 14:09:38 | 16:20:11 |  BUS  |    Sne   | 063     | 86061 84932       | WALK 13:22 - 86061 - BUS 063 14:24 16:20 - 84932 - WALK 0:11

Not all information is saved/retrieved from the travelSearch-expected-results.csv file, so the value for there fields are just a '-' in the table.

In the ItineraryResultMapper there is a map of agencies/abbreviations used in the debug logging. You may want to add your agencies to get pretty results.

Running the HTTP Server

We have copied the OTP GUI, and set up a minimalistic HTTP Server to work with the Speed Test. We rarely use it, so notify us if it does not work.

TODO TGR