timeinfo tracking in map matching #2712

kevinkreiser · 2020-12-07T16:29:27Z

Currently map matching (either fast or robust versions) dont use the TimeInfo object to track time along the route (once the match is found). This pr adds that tracking into both versions. It also makes some changes in the fast map matcher to prepare for supporting multileg matches. Right now i've gone a little too far with the multi leg support that sometimes the edge walk fails. So i'll either need to finish support for that or rollback some of the changes.

kevinkreiser · 2021-01-25T02:45:54Z

src/loki/height_action.cc

-namespace {
-PointLL to_ll(const valhalla::Location& l) {
-  return PointLL{l.ll().lng(), l.ll().lat()};
-}
-void from_ll(valhalla::Location* l, const PointLL& p) {
-  l->mutable_ll()->set_lat(p.lat());
-  l->mutable_ll()->set_lng(p.lng());
-}
-} // namespace


this existed in the header above so i did this little clean up

kevinkreiser · 2021-01-25T02:46:58Z

src/thor/map_matcher.cc

-  uint32_t origin_epoch = compute_origin_epoch(edge_segments, matcher, options);
-  std::string date_time = options.shape(0).has_date_time() ? options.shape(0).date_time() : "";
+  valhalla::baldr::DateTime::tz_sys_info_cache_t tz_cache;
+  auto time_info = init_time_info(edge_segments, matcher, options, &tz_cache);


this is the main point of this pr. we dont manually track second of week or the timezone etc. we do it with the time_info object as all the other time based algorithms do it

kevinkreiser · 2021-01-25T02:47:58Z

src/thor/route_matcher.cc

-    }
-  }
+  valhalla::baldr::DateTime::tz_sys_info_cache_t tz_cache;
+  auto time_info = init_time_info(reader, options, &tz_cache);


so for the shape walking map matcher (ie the faster less robust one) we also switch to using time_info

kevinkreiser · 2021-01-25T02:49:15Z

src/thor/route_matcher.cc

                            valhalla::Options& options,
-                            std::vector<PathInfo>& path_infos) {
+                            std::vector<std::vector<PathInfo>>& legs) {


i started refactoring this method to support multileg routes but there needs to be a little bit more refactoring in the guts of these algorithm to make it possible. i started trying it but it was getting too hair so i aborted that for now

kevinkreiser · 2021-01-25T02:50:18Z

src/thor/trace_route_action.cc

-                                *options.mutable_locations()->rbegin(),
-                                std::list<valhalla::Location>{}, leg, {"edge_walk"}, interrupt);
-  } else {
+  std::vector<std::vector<PathInfo>> legs;


yeah so i refactored this so that when we finally get multileg support in the fast map matching algorithm this code will already be done, but as it stands today its still not supported

gknisely · 2021-01-27T16:56:27Z

@kevinkreiser do we have existing tests already or do we need some

kevinkreiser · 2021-01-27T17:00:41Z

@gknisely there are existing ones that cover depart_at but time invariant doesnt exist so far as i know. that behavior though is all wrapped into the timeinfo object. do you want me to add a few simple test cases just in case. should be quick

gknisely · 2021-01-27T17:33:37Z

@gknisely there are existing ones that cover depart_at but time invariant doesnt exist so far as i know. that behavior though is all wrapped into the timeinfo object. do you want me to add a few simple test cases just in case. should be quick

@kevinkreiser yes. just to be safe

…feretn t ypes of traffic data and different times of time dependence

…edep_mapmatch

kevinkreiser · 2021-01-29T04:06:44Z

src/meili/candidate_search.cc

@@ -44,17 +44,17 @@ CandidateCollector::WithinSquaredDistance(const midgard::PointLL& location,
      continue;
    }

-    // Get the edge and its opposing edge. Transition edges are not
-    // allowed so we do not need to check node levels.
-    const auto opp_edgeid = reader_.GetOpposingEdgeId(edgeid, tile);


this was a bug. when you get the opposing id with the tile of the regular edge it can flip modify the tile to be the wrong tile which we use again to refer to the original edge. the code was throwing an exception, getting caught and failing the map match. this would happen on all map match input points which have radii large enough to intersect an edge that crosses a tile boundary

cc @purew @mandeepsandhu @mookerji

kevinkreiser · 2021-01-29T04:07:15Z

src/mjolnir/graphtilebuilder.cc

@@ -1021,11 +1021,8 @@ void GraphTileBuilder::AddBins(const std::string& tile_dir,

 // Add a predicted speed profile for a directed edge.
 void GraphTileBuilder::AddPredictedSpeed(const uint32_t idx,
-                                         const std::vector<int16_t>& profile,
+                                         const std::array<int16_t, kCoefficientCount>& coefficients,


it was stupid to take a vector when we wanted the size of the thing to be exactly a certain length. std::array does this. so lets use that instead

kevinkreiser · 2021-01-29T04:08:40Z

test/gurka/CMakeLists.txt

+    set_target_properties(${TESTNAME} PROPERTIES FOLDER "Tests")
+    target_compile_definitions(${TESTNAME} PRIVATE
+      VALHALLA_SOURCE_DIR="${VALHALLA_SOURCE_DIR}/"
+      VALHALLA_BUILD_DIR="${VALHALLA_BUILD_DIR}/")


need this to reference properly the timezone sqlite that is built for use in timedependent tests

kevinkreiser · 2021-01-29T04:09:30Z

test/gurka/test_route.cc

@@ -247,7 +247,7 @@ TEST(AutoDataFix, deprecation) {
 class AlgorithmTest : public ::testing::Test {
 protected:
  static gurka::map map;
-  static uint32_t current, historical, constrained;
+  static uint32_t current, historical, constrained, freeflow;


i used this test as a model for what i did in the map matching test. they are basically idential and return identical results

kevinkreiser · 2021-01-29T04:10:12Z

test/test.h

@@ -84,7 +84,7 @@ using LiveTrafficCustomize = std::function<void(valhalla::baldr::GraphReader&,
 void customize_live_traffic_data(const boost::property_tree::ptree& config,
                                 const LiveTrafficCustomize& setter_cb);

-using HistoricalTrafficCustomize = std::function<std::vector<int16_t>(DirectedEdge&)>;
+using HistoricalTrafficCustomize = std::function<std::array<float, kBucketsPerWeek>(DirectedEdge&)>;


this is way more useful if you give it the raw speeds and let it convert it to the format stored in the tiles

kevinkreiser · 2021-01-29T04:10:56Z

test/gurka/test_match.cc

+}
+
+// map matching currently only allows forward time tracking and invariant so types 0, 1, 3.
+// type 2 arrive_by is not yet supported


we check all the different types of su pported timedependent map matching as well as all the different sources of speed data

kevinkreiser added 4 commits December 5, 2020 19:56

use baldr::TimeInfo in meili map matching FormPath

59c7486

simplify edge walking and prepare it to allow break location types

9f42d0c

more prep for legs

a4a203c

merge conflicts

3df14ba

kevinkreiser marked this pull request as ready for review January 23, 2021 16:47

Merge branch 'master' into kk_timedep_mapmatch

6953cc2

kevinkreiser requested a review from genadz January 25, 2021 02:33

changelog

c8b7cec

kevinkreiser commented Jan 25, 2021

View reviewed changes

cleanup

28563a4

kevinkreiser requested a review from mookerji January 25, 2021 17:32

genadz previously approved these changes Jan 27, 2021

View reviewed changes

kevinkreiser added 2 commits January 27, 2021 08:40

Merge branch 'master' into kk_timedep_mapmatch

47b34b7

Update CHANGELOG.md

3550100

kevinkreiser dismissed genadz’s stale review via 3550100 January 27, 2021 13:42

kevinkreiser added 3 commits January 27, 2021 09:23

Merge branch 'master' into kk_timedep_mapmatch

1db7671

already have the baldr namespace

893505e

Merge branch 'master' into kk_timedep_mapmatch

d69d460

genadz previously approved these changes Jan 27, 2021

View reviewed changes

kevinkreiser added 3 commits January 28, 2021 22:33

fix another case of wrong tile wrong edge bug

59cbaf0

increase map matchign test coverage cases involving routing using dif…

8e58b2d

…feretn t ypes of traffic data and different times of time dependence

Merge remote-tracking branch 'origin/kk_timedep_mapmatch' into kk_tim…

edd75c9

…edep_mapmatch

kevinkreiser dismissed genadz’s stale review via edd75c9 January 29, 2021 03:37

Merge branch 'master' into kk_timedep_mapmatch

b9cc041

kevinkreiser commented Jan 29, 2021

View reviewed changes

gknisely self-requested a review January 29, 2021 13:52

gknisely approved these changes Jan 29, 2021

View reviewed changes

kevinkreiser merged commit ce39088 into master Jan 29, 2021

kevinkreiser mentioned this pull request Feb 2, 2021

more test fixes for historical traffic regression #2829

Merged

kevinkreiser deleted the kk_timedep_mapmatch branch January 5, 2022 03:37

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

timeinfo tracking in map matching #2712

timeinfo tracking in map matching #2712

kevinkreiser commented Dec 7, 2020

kevinkreiser Jan 25, 2021

kevinkreiser Jan 25, 2021

kevinkreiser Jan 25, 2021

kevinkreiser Jan 25, 2021

kevinkreiser Jan 25, 2021

gknisely commented Jan 27, 2021

kevinkreiser commented Jan 27, 2021

gknisely commented Jan 27, 2021

kevinkreiser Jan 29, 2021

kevinkreiser Jan 29, 2021

kevinkreiser Jan 29, 2021

kevinkreiser Jan 29, 2021

kevinkreiser Jan 29, 2021

kevinkreiser Jan 29, 2021

kevinkreiser Jan 29, 2021

timeinfo tracking in map matching #2712

timeinfo tracking in map matching #2712

Conversation

kevinkreiser commented Dec 7, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gknisely commented Jan 27, 2021

kevinkreiser commented Jan 27, 2021

gknisely commented Jan 27, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment