Summary Doesnt Match Accumulated Maneuvers #2195

kevinkreiser · 2020-01-30T23:15:39Z

as the title says, the summary time information doesnt match the time you accumulate from the maneuver object. even for relatively short routes we seem to be accumulating rounding error in the summary or truncating on the maneuvers

so it turns out that we do a bunch of truncation in odin when we create the directions object. so ive updated all of those to double precision. additionally i added 3 decimal places of precision to the output in the json. what do people think of that. millisecond resolution. perhaps we only need 1 decimal place but the trick is that if you do that you accumulate error faster where you have a lot of maneuvers.

…ot add up

kevinkreiser · 2020-02-03T21:36:58Z

test/summary.cc

+      R"({"locations":[{"lat":52.048267,"lon":5.074825},{"lat":52.114622,"lon":5.131816}],"costing":"auto"})";
+  auto response = tester.test(request);
+
+  // loop over all routes all legs


first part of the test is to loop over the internal protobuf route representations, both the path and the directions objects and compare that they all add up to the same thing

kevinkreiser · 2020-02-03T21:37:33Z

test/summary.cc

+    ++trip_route;
+  }
+
+  // get the json


then we loop over valhalla flavored json and we have to give 10ms of tolerance because in the end we do rounding when we serialize the values to json

kevinkreiser · 2020-02-03T21:37:50Z

test/summary.cc

+    EXPECT_NEAR(accumulated_time, leg["summary"].GetObject()["time"].GetDouble(), .01);
+  }
+
+  // get the osrm json


finally we loop over the osrm json to check that that also adds up properly

kevinkreiser · 2020-02-03T21:39:06Z

src/tyr/route_serializer_osrm.cc

-  route->emplace("duration", json::fp_t{duration, 1});
+  distance *= imperial ? 1609.344 : 1000.0;
+  route->emplace("distance", json::fp_t{distance, 3});
+  route->emplace("duration", json::fp_t{duration, 3});


are milliseconds too fine a measure?

👍 milliseconds makes sense to me to minimize accumulated error

kevinkreiser · 2020-02-03T21:44:15Z

scripts/valhalla_build_config

@@ -17,6 +17,7 @@ config = {
    'user_agent': optional(str),
    'tile_url': optional(str),
    'tile_url_gz': optional(bool),
+    'concurrency': optional(int),


unrelated change, but i realized it was an undocumented feature in tile building, so i added it

kevinkreiser · 2020-02-03T21:45:02Z

valhalla/odin/enhancedtrippath.h

@@ -557,7 +557,7 @@ class EnhancedTripLeg_Node {
    return mutable_node_->type();
  }

-  uint32_t elapsed_time() const {
+  double elapsed_time() const {
    return mutable_node_->elapsed_time();


technically didnt need to do this as its only for debug output but, it didnt feel right to truncate the number

dgearhart · 2020-02-03T21:45:32Z

proto/directions.proto

@@ -142,7 +142,7 @@ message DirectionsLeg {
    optional string text_instruction = 2;                    // text instruction
    repeated StreetName street_name = 3;                     // street names
    optional float length = 4;                               // kilometers or miles based on units


should length be switched to double too?

i thought about it but it seemed somewhat unrelated to the PR at hand. i could expand the pr to test lengths as well. is that the concensus?

I think length can be handled in a separate PR. I was just a bit confused as I saw spots in the code where you change the precision of distance in the response.

i was thinking to be consistent since distance was returned as double

i changed the decimal places in json but not the internal precision (which is float). we should change internal distance to be double but at the moment that is a larger change

kevinkreiser · 2020-02-03T21:54:53Z

well oops. i didnt realize that other tests were relying on this (even though i wrote them).

dgearhart

if tests pass then 🚢

kevinkreiser · 2020-02-03T23:29:58Z

all builds pass. going to update changelog and merge

* master: Clarify cmake flag preconditions. Fix coverage flag. Include appropriate license header. Disable -Werror in CI build. Remove -Wconversion Update README and CI configuration. Revert coverage flags back to being PUBLIC. Remove stray comment. Refactored flags to use a function; fixed valhalla_module target; flag for Werror Adds a cmake module for libcxx flag macros. Stop impact, PH, turn channel speeds, and tagged speed updates (#2198) add isochrone snapped center coordinates to response (#2111) Summary Doesnt Match Accumulated Maneuvers (#2195)

danpat · 2020-02-10T17:11:09Z

src/tyr/route_serializer_osrm.cc

-  distance *= imperial ? 1609.34f : 1000.0f;
-  route->emplace("distance", json::fp_t{distance, 1});
-  route->emplace("duration", json::fp_t{duration, 1});
+  distance *= imperial ? 1609.344 : 1000.0;


Can you factor this out into a couple of constants somewhere? Do these already exist somewhere that could be reused?

there are distance constants in midgard/constants.h.
constexpr float kKmPerMile = 1.60934f;
Just need to add kMetersPerMile

kevinkreiser added 2 commits January 30, 2020 18:13

make a new failing test showing summary and individual maneuvers do n…

0283589

…ot add up

up the precision of time information in odin and json output

291d85f

kevinkreiser marked this pull request as ready for review February 3, 2020 21:30

kevinkreiser requested review from dgearhart and danpaz February 3, 2020 21:30

kevinkreiser commented Feb 3, 2020

View reviewed changes

dgearhart reviewed Feb 3, 2020

View reviewed changes

kevinkreiser added 2 commits February 3, 2020 17:40

add tolerance to length now that its got more decimal places

f69f342

Merge branch 'master' into kk_time

b549bd7

dgearhart previously approved these changes Feb 3, 2020

View reviewed changes

danpaz previously approved these changes Feb 3, 2020

View reviewed changes

Update CHANGELOG.md

fa533b7

kevinkreiser dismissed stale reviews from danpaz and dgearhart via fa533b7 February 3, 2020 23:32

kevinkreiser merged commit 32b8c04 into master Feb 3, 2020

purew deleted the kk_time branch February 4, 2020 21:46

danpat reviewed Feb 10, 2020

View reviewed changes

kevinkreiser mentioned this pull request Feb 10, 2020

missed leg weight to 3 decimal places #2227

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Summary Doesnt Match Accumulated Maneuvers #2195

Summary Doesnt Match Accumulated Maneuvers #2195

kevinkreiser commented Jan 30, 2020 •

edited

Loading

kevinkreiser Feb 3, 2020

kevinkreiser Feb 3, 2020

kevinkreiser Feb 3, 2020

kevinkreiser Feb 3, 2020

danpaz Feb 3, 2020

kevinkreiser Feb 3, 2020

kevinkreiser Feb 3, 2020

dgearhart Feb 3, 2020

kevinkreiser Feb 3, 2020

danpaz Feb 3, 2020

dgearhart Feb 3, 2020

kevinkreiser Feb 3, 2020

kevinkreiser commented Feb 3, 2020

dgearhart left a comment

kevinkreiser commented Feb 3, 2020

danpat Feb 10, 2020

dnesbitt61 Feb 10, 2020

Summary Doesnt Match Accumulated Maneuvers #2195

Summary Doesnt Match Accumulated Maneuvers #2195

Conversation

kevinkreiser commented Jan 30, 2020 • edited Loading

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

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

kevinkreiser commented Feb 3, 2020

dgearhart left a comment

Choose a reason for hiding this comment

kevinkreiser commented Feb 3, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kevinkreiser commented Jan 30, 2020 •

edited

Loading