playing with processing modifications (#183)

* WIP set up plumbing for calculating building info in the Raw layer and using it later for #154

* WIP playing with processing modifications

* experiment with some possibilities for tag processing

* fix an embarassing typo

* enable busses in work-home traffic

* refactor, add multiuse

* seed more parking in Kraków

* parse integers properly - thanks for a help!

* rebalance generated trips

* add hack providing some background traffic

more realistic traffic

even mess realistic people

* attempt to further reduce parking deficit

* add TODO

Co-authored-by: Dustin Carlino <dabreegster@gmail.com>

Cargo.lock

@@ -1814,6 +1814,8 @@ dependencies = [
  "geom 0.1.0",
  "nbez 0.1.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "petgraph 0.5.0 (registry+https://github.com/rust-lang/crates.io-index)",
+ "rand 0.7.3 (registry+https://github.com/rust-lang/crates.io-index)",
+ "rand_xorshift 0.2.0 (registry+https://github.com/rust-lang/crates.io-index)",
  "seattle_traffic_signals 0.1.0 (git+https://github.com/dabreegster/seattle_traffic_signals)",
  "serde 1.0.110 (registry+https://github.com/rust-lang/crates.io-index)",
  "thread_local 1.0.1 (registry+https://github.com/rust-lang/crates.io-index)",

convert_osm/src/lib.rs

@@ -39,7 +39,7 @@ pub enum OnstreetParking {
     Blockface(String),
     // If OSM data is missing, then infer parking lanes on some percentage of
     // "highway=residential" roads.
-    SomeResidential {
+    SomeAdditionalWhereNoData {
         // [0, 100]
         pct: usize,
     },
@@ -86,11 +86,15 @@ pub fn convert(opts: Options, timer: &mut abstutil::Timer) -> RawMap {
         OnstreetParking::Blockface(ref path) => {
             use_parking_hints(&mut map, path.clone(), timer);
         }
-        OnstreetParking::SomeResidential { pct } => {
+        OnstreetParking::SomeAdditionalWhereNoData { pct } => {
             let pct = pct as i64;
             for (id, r) in map.roads.iter_mut() {
                 if r.osm_tags.contains_key(osm::INFERRED_PARKING)
-                    && r.osm_tags.get(osm::HIGHWAY) == Some(&"residential".to_string())
+                    && (
+                        r.osm_tags.get(osm::HIGHWAY) == Some(&"residential".to_string())
+                        || r.osm_tags.get(osm::HIGHWAY) == Some(&"tertiary".to_string())
+                        || r.osm_tags.get(osm::HIGHWAY) == Some(&"living_street".to_string())
+                       ) 
                     && id.osm_way_id % 100 <= pct
                 {
                     if r.osm_tags.get("oneway") == Some(&"yes".to_string()) {

importer/src/krakow.rs

@@ -31,9 +31,10 @@ pub fn osm_to_raw(name: &str) {
                 bikes_can_use_bus_lanes: false,
             },
 
-            onstreet_parking: convert_osm::OnstreetParking::SomeResidential { pct: 50 },
+            onstreet_parking: convert_osm::OnstreetParking::SomeAdditionalWhereNoData { pct: 90 },
             public_offstreet_parking: convert_osm::PublicOffstreetParking::None,
-            private_offstreet_parking: convert_osm::PrivateOffstreetParking::FixedPerBldg(1),
+            private_offstreet_parking: convert_osm::PrivateOffstreetParking::FixedPerBldg(3), // TODO: support amenity=parking_entrance
+            // TODO: investigate why some many buildings drop their private parkings
             elevation: None,
         },
         &mut abstutil::Timer::throwaway(),

map_model/Cargo.toml

@@ -15,3 +15,5 @@ petgraph = "0.5.0"
 serde = "1.0.110"
 thread_local = "1.0.1"
 seattle_traffic_signals = { git = "https://github.com/dabreegster/seattle_traffic_signals" }
+rand = "0.7.0"
+rand_xorshift = "0.2.0"

map_model/src/lib.rs

@@ -16,7 +16,9 @@ pub use crate::edits::{
 pub use crate::make::initial::lane_specs::RoadSpec;
 pub use crate::map::MapConfig;
 pub use crate::objects::area::{Area, AreaID, AreaType};
-pub use crate::objects::building::{Building, BuildingID, FrontPath, OffstreetParking};
+pub use crate::objects::building::{
+    Building, BuildingID, BuildingType, FrontPath, OffstreetParking,
+};
 pub use crate::objects::bus_stop::{BusRoute, BusRouteID, BusStop, BusStopID};
 pub use crate::objects::intersection::{Intersection, IntersectionID, IntersectionType};
 pub use crate::objects::lane::{

map_model/src/make/buildings.rs

@@ -1,12 +1,14 @@
 use crate::make::match_points_to_lanes;
 use crate::raw::{OriginalBuilding, RawBuilding, RawParkingLot};
 use crate::{
-    osm, Building, BuildingID, FrontPath, LaneID, LaneType, Map, OffstreetParking, ParkingLot,
+    osm, Building, BuildingID, BuildingType, FrontPath, LaneID, LaneType, Map, OffstreetParking,
-    ParkingLotID, Position, NORMAL_LANE_THICKNESS, PARKING_LOT_SPOT_LENGTH,
+    ParkingLot, ParkingLotID, Position, NORMAL_LANE_THICKNESS, PARKING_LOT_SPOT_LENGTH,
 };
 use abstutil::Timer;
 use geom::{Angle, Distance, FindClosest, HashablePt2D, Line, PolyLine, Polygon, Pt2D, Ring};
-use std::collections::{BTreeMap, HashSet};
+use std::collections::{BTreeMap, BTreeSet, HashSet};
+use rand::{Rng, SeedableRng};
+use rand_xorshift::XorShiftRng;
 
 pub fn make_all_buildings(
     input: &BTreeMap<OriginalBuilding, RawBuilding>,
@@ -57,6 +59,7 @@ pub fn make_all_buildings(
             };
 
             let id = BuildingID(results.len());
+            let mut rng = XorShiftRng::seed_from_u64(orig_id.osm_way_id as u64);
             let mut bldg = Building {
                 id,
                 polygon: b.polygon.clone(),
@@ -70,6 +73,7 @@ pub fn make_all_buildings(
                 amenities: b.amenities.clone(),
                 parking: None,
                 label_center: b.polygon.polylabel(),
+                bldg_type: classify_bldg(&b.osm_tags, &b.amenities, b.polygon.area(), &mut rng),
             };
 
             // Can this building have a driveway? If it's not next to a driving lane, then no.
@@ -354,3 +358,73 @@ fn line_valid(
 
     true
 }
+
+fn classify_bldg(
+    tags: &BTreeMap<String, String>,
+    amenities: &BTreeSet<(String, String)>,
+    area_sq_meters: f64,
+    rng: &mut rand_xorshift::XorShiftRng,
+) -> BuildingType {
+    // used: top values from https://taginfo.openstreetmap.org/keys/building#values (>100k uses)
+    let tags = Tags(tags);
+
+    let mut commercial = false;
+    let mut workers = 0;
+ 
+    // These are (name, amenity type) pairs, produced by get_bldg_amenities in
+    // convert_osm/src/osm_reader.rs.
+    if !amenities.is_empty() {
+        commercial = true;
+    }
+
+    if tags.is("ruins", "yes") {
+        if commercial {
+            return BuildingType::Commercial;
+        }
+        return BuildingType::Empty;
+    }
+
+    if tags.is_any("building", vec!["office", "industrial", "commercial", "retail", "warehouse", "civic", "public"]) {
+        return BuildingType::Commercial;
+    } else if tags.is_any("building", vec!["school", "university", "construction", "church"]) {
+        // TODO: special handling in future
+        return BuildingType::Empty;
+    } else if tags.is_any("building", vec!["garage", "garages", "shed", "roof", "greenhouse", "farm_auxiliary", "barn", "service"]) {
+        return BuildingType::Empty;
+    } else if tags.is_any("building", vec!["house", "detached", "semidetached_house", "farm"]) {
+        workers = rng.gen_range(0, 3);
+    } else if tags.is_any("building", vec!["hut", "static_caravan", "cabin"]) {
+        workers = rng.gen_range(0, 2);
+    } else if tags.is_any("building", vec!["apartments", "terrace", "residential"]) {
+        let levels = tags.0.get("building:levels").and_then(|x| x.parse::<usize>().ok()).unwrap_or(1);
+        // TODO is it worth using height or building:height as an alternative if not tagged?
+        // 1 person per 10 square meters
+        let residents = (levels as f64 * area_sq_meters / 10.0) as usize;
+        workers = (residents / 3) as usize;
+    } else {
+        workers = rng.gen_range(0, 2);
+    }
+    if commercial {
+        if workers > 0 {
+            return BuildingType::ResidentialCommercial(workers);
+        }
+        return BuildingType::Commercial;
+    }
+    return BuildingType::Residential(workers);
+}
+
+// TODO Refactor with lane_specs
+struct Tags<'a>(&'a BTreeMap<String, String>);
+impl<'a> Tags<'a> {
+    fn is(&self, k: &str, v: &str) -> bool {
+        self.0.get(k) == Some(&v.to_string())
+    }
+
+    fn is_any(&self, k: &str, values: Vec<&str>) -> bool {
+        if let Some(v) = self.0.get(k) {
+            values.contains(&v.as_ref())
+        } else {
+            false
+        }
+    }
+}

map_model/src/objects/building.rs

@@ -50,6 +50,7 @@ pub struct Building {
     // Where a text label should be centered to have the best chances of being contained within the
     // polygon.
     pub label_center: Pt2D,
+    // TODO Might fold these into BuildingType::Commercial
     // (Name, amenity)
     pub amenities: BTreeSet<(String, String)>,
 
@@ -57,6 +58,16 @@ pub struct Building {
     // Every building can't have OffstreetParking, because the nearest usable driving lane (not in
     // a parking blackhole) might be far away
     pub parking: Option<OffstreetParking>,
+    pub bldg_type: BuildingType,
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+pub enum BuildingType {
+    // An estimated number of residents
+    Residential(usize),
+    ResidentialCommercial(usize),
+    Commercial,
+    Empty,
 }
 
 impl Building {

sim/src/make/generator.rs

@@ -4,7 +4,7 @@ use crate::{
 };
 use abstutil::Timer;
 use geom::{Distance, Duration, Time};
-use map_model::{BuildingID, DirectedRoadID, Map, PathConstraints, PathRequest};
+use map_model::{BuildingID, BuildingType, DirectedRoadID, Map, PathConstraints, PathRequest};
 use rand::seq::SliceRandom;
 use rand::Rng;
 use rand_xorshift::XorShiftRng;
@@ -418,29 +418,41 @@ fn rand_time(rng: &mut XorShiftRng, low: Time, high: Time) -> Time {
 impl ScenarioGenerator {
     // Designed in https://github.com/dabreegster/abstreet/issues/154
     pub fn proletariat_robot(map: &Map, rng: &mut XorShiftRng, timer: &mut Timer) -> Scenario {
-        // First classify buildings into residences (with a number of people) or workplaces. No
-        // mixed-use yet; every building is one or the other.
         let mut residences: Vec<(BuildingID, usize)> = Vec::new();
         let mut workplaces: Vec<BuildingID> = Vec::new();
         let mut total_ppl = 0;
         for b in map.all_buildings() {
-            // These are scraped from OSM "shop" and "amenity" tags.
+            match b.bldg_type {
-            if b.amenities.is_empty() {
+                BuildingType::Residential(num_ppl) => {
-                // TODO Guess number of residences based on OSM tags.
+                    residences.push((b.id, num_ppl));
-                let num_ppl = rng.gen_range(1, 5);
+                    total_ppl += num_ppl;
+                }
+                BuildingType::ResidentialCommercial(num_ppl) => {
                     residences.push((b.id, num_ppl));
                     total_ppl += num_ppl;
-            } else {
                     workplaces.push(b.id);
                 }
+                BuildingType::Commercial => {
+                    workplaces.push(b.id);
+                }
+                BuildingType::Empty => {}
+            }
         }
 
         let mut s = Scenario::empty(map, "random people going to/from work");
+        s.only_seed_buses = None;
         timer.start_iter("create people", total_ppl);
         for (home, num_ppl) in residences {
             for _ in 0..num_ppl {
                 timer.next();
                 // Make a person going from their home to a random workplace, then back again later.
+                
+                // TODO refactor
+                // function or associated item not found in `rand_xorshift::XorShiftRng`
+                // so why it works in buildings.rs?
+                //let mut deterministicRng = XorShiftRng::seed_from_u64(home.0 as u64);
+                //let work = *workplaces.choose(deterministicRng).unwrap();
+
                 let work = *workplaces.choose(rng).unwrap();
                 // Decide mode based on walking distance.
                 let dist = if let Some(path) = map.pathfind(PathRequest {
@@ -454,14 +466,33 @@ impl ScenarioGenerator {
                     // this person.
                     continue;
                 };
-                // Trips over 2 miles will drive 90% of the time, the other 10% will attempt
+                if home.0 == work.0 {
-                // transit (falling back to a very long walk).
+                    // working and living in the same building
+                    continue;
+                }
+                // Longer trips will mostly drive of the time, remaining will attempt
+                // transit (falling back to a very long walk), with some small number of people cycling.
                 // TODO Make this probabilistic
-                let mode = if dist < Distance::miles(1.0) {
+                // TODO - do not select based on distance but select one that is fastest/best in the given situation
+                // excellent bus connection / plenty of parking / cycleways / suitable rail connection
+                // all strongly influence selected mode of transport, distance is not the sole influence
+                let mode = if dist < Distance::miles(0.5) {
                     TripMode::Walk
                 } else if dist < Distance::miles(2.0) {
+                    if rng.gen_bool(0.3) {
+                        // 30%
+                        TripMode::Transit
+                    } else if rng.gen_bool(0.8) {
+                        // 0.7 * 0.8 = 56%
+                        TripMode::Drive
+                    } else {
+                        // 14%
                         TripMode::Bike
-                } else if rng.gen_bool(0.9) {
+                    }
+                } else if rng.gen_bool(0.005) {
+                    // low chance for really, really dedicated cyclists
+                    TripMode::Bike
+                } else if rng.gen_bool(0.7) {
                     TripMode::Drive
                 } else {
                     TripMode::Transit
@@ -470,17 +501,32 @@ impl ScenarioGenerator {
                 // TODO This will cause a single morning and afternoon rush. Outside of these times,
                 // it'll be really quiet. Probably want a normal distribution centered around these
                 // peak times, but with a long tail.
-                let depart_am = rand_time(
+                let mut depart_am = rand_time(
                     rng,
                     Time::START_OF_DAY + Duration::hours(7),
                     Time::START_OF_DAY + Duration::hours(10),
                 );
-                let depart_pm = rand_time(
+                let mut depart_pm = rand_time(
                     rng,
                     Time::START_OF_DAY + Duration::hours(17),
                     Time::START_OF_DAY + Duration::hours(19),
                 );
 
+                if rng.gen_bool(0.1) {
+                    // hacky hack to get some background traffic
+                    // TODO - avoid mutable variable
+                    depart_am = rand_time(
+                        rng,
+                        Time::START_OF_DAY + Duration::hours(0),
+                        Time::START_OF_DAY + Duration::hours(12),
+                    );
+                    depart_pm = rand_time(
+                        rng,
+                        Time::START_OF_DAY + Duration::hours(12),
+                        Time::START_OF_DAY + Duration::hours(24),
+                    );
+                }
+
                 let (goto_work, return_home) = match (
                     SpawnTrip::new(
                         TripEndpoint::Bldg(home),