public GraphIndex(Graph graph) {
LOG.info("Indexing graph...");
for (String feedId : graph.getFeedIds()) {
for (Agency agency : graph.getAgencies(feedId)) {
Map<String, Agency> agencyForId = agenciesForFeedId.getOrDefault(feedId, new HashMap<>());
agencyForId.put(agency.getId(), agency);
this.agenciesForFeedId.put(feedId, agencyForId);
}
}
Collection<Edge> edges = graph.getEdges();
/* We will keep a separate set of all vertices in case some have the same label.
* Maybe we should just guarantee unique labels. */
Set<Vertex> vertices = Sets.newHashSet();
for (Edge edge : edges) {
vertices.add(edge.getFromVertex());
vertices.add(edge.getToVertex());
if (edge instanceof TablePatternEdge) {
TablePatternEdge patternEdge = (TablePatternEdge) edge;
TripPattern pattern = patternEdge.getPattern();
patternForId.put(pattern.code, pattern);
}
}
for (Vertex vertex : vertices) {
vertexForId.put(vertex.getLabel(), vertex);
if (vertex instanceof TransitStop) {
TransitStop transitStop = (TransitStop) vertex;
Stop stop = transitStop.getStop();
stopForId.put(stop.getId(), stop);
stopVertexForStop.put(stop, transitStop);
stopsForParentStation.put(stop.getParentStation(), stop);
}
}
for (TransitStop stopVertex : stopVertexForStop.values()) {
Envelope envelope = new Envelope(stopVertex.getCoordinate());
stopSpatialIndex.insert(envelope, stopVertex);
}
for (TripPattern pattern : patternForId.values()) {
patternsForFeedId.put(pattern.getFeedId(), pattern);
patternsForRoute.put(pattern.route, pattern);
for (Trip trip : pattern.getTrips()) {
patternForTrip.put(trip, pattern);
tripForId.put(trip.getId(), trip);
}
for (Stop stop : pattern.getStops()) {
patternsForStop.put(stop, pattern);
}
}
for (Route route : patternsForRoute.asMap().keySet()) {
routeForId.put(route.getId(), route);
}
// Copy these two service indexes from the graph until we have better ones.
calendarService = graph.getCalendarService();
serviceCodes = graph.serviceCodes;
this.graph = graph;
LOG.info("Done indexing graph.");
}
/**
* Stop clustering is slow to perform and only used in profile routing for the moment. Therefore
* it is not done automatically, and any method requiring stop clusters should call this method to
* ensure that the necessary indexes are lazy-initialized.
*/
public synchronized void clusterStopsAsNeeded() {
if (stopClusterSpatialIndex == null) {
clusterStops();
LOG.info("Creating a spatial index for stop clusters.");
stopClusterSpatialIndex = new HashGridSpatialIndex<StopCluster>();
for (StopCluster cluster : stopClusterForId.values()) {
Envelope envelope = new Envelope(new Coordinate(cluster.lon, cluster.lat));
stopClusterSpatialIndex.insert(envelope, cluster);
}
}
}
/**
* Find transfer candidates for profile routing. TODO replace with an on-street search using the
* existing profile router functions.
*/
public Map<StopCluster, Double> findNearbyStopClusters(StopCluster sc, double radius) {
Map<StopCluster, Double> ret = Maps.newHashMap();
Envelope env = new Envelope(new Coordinate(sc.lon, sc.lat));
env.expandBy(
SphericalDistanceLibrary.metersToLonDegrees(radius, sc.lat),
SphericalDistanceLibrary.metersToDegrees(radius));
for (StopCluster cluster : stopClusterSpatialIndex.query(env)) {
// TODO this should account for area-like nature of clusters. Use size of bounding boxes.
double distance = SphericalDistanceLibrary.distance(sc.lat, sc.lon, cluster.lat, cluster.lon);
if (distance < radius) ret.put(cluster, distance);
}
return ret;
}
/**
* FIXME OBA parentStation field is a string, not an AgencyAndId, so it has no agency/feed scope
* But the DC regional graph has no parent stations pre-defined, so no use dealing with them for
* now. However Trimet stops have "landmark" or Transit Center parent stations, so we don't use
* the parent stop field.
*
* <p>Ideally in the future stop clusters will replicate and/or share implementation with GTFS
* parent stations.
*
* <p>We can't use a similarity comparison, we need exact matches. This is because many street
* names differ by only one letter or number, e.g. 34th and 35th or Avenue A and Avenue B.
* Therefore normalizing the names before the comparison is essential. The agency must provide
* either parent station information or a well thought out stop naming scheme to cluster stops --
* no guessing is reasonable without that information.
*/
public void clusterStops() {
int psIdx = 0; // unique index for next parent stop
LOG.info("Clustering stops by geographic proximity and name...");
// Each stop without a cluster will greedily claim other stops without clusters.
for (Stop s0 : stopForId.values()) {
if (stopClusterForStop.containsKey(s0))
continue; // skip stops that have already been claimed by a cluster
String s0normalizedName = StopNameNormalizer.normalize(s0.getName());
StopCluster cluster = new StopCluster(String.format("C%03d", psIdx++), s0normalizedName);
// LOG.info("stop {}", s0normalizedName);
// No need to explicitly add s0 to the cluster. It will be found in the spatial index query
// below.
Envelope env = new Envelope(new Coordinate(s0.getLon(), s0.getLat()));
env.expandBy(
SphericalDistanceLibrary.metersToLonDegrees(CLUSTER_RADIUS, s0.getLat()),
SphericalDistanceLibrary.metersToDegrees(CLUSTER_RADIUS));
for (TransitStop ts1 : stopSpatialIndex.query(env)) {
Stop s1 = ts1.getStop();
double geoDistance =
SphericalDistanceLibrary.fastDistance(
s0.getLat(), s0.getLon(), s1.getLat(), s1.getLon());
if (geoDistance < CLUSTER_RADIUS) {
String s1normalizedName = StopNameNormalizer.normalize(s1.getName());
// LOG.info(" --> {}", s1normalizedName);
// LOG.info(" geodist {} stringdist {}", geoDistance, stringDistance);
if (s1normalizedName.equals(s0normalizedName)) {
// Create a bidirectional relationship between the stop and its cluster
cluster.children.add(s1);
stopClusterForStop.put(s1, cluster);
}
}
}
cluster.computeCenter();
stopClusterForId.put(cluster.id, cluster);
}
// LOG.info("Done clustering stops.");
// for (StopCluster cluster : stopClusterForId.values()) {
// LOG.info("{} at {} {}", cluster.name, cluster.lat, cluster.lon);
// for (Stop stop : cluster.children) {
// LOG.info(" {}", stop.getName());
// }
// }
}