Search in sources :

Example 1 with EuclideanRemainingWeightHeuristic

use of org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic in project OpenTripPlanner by opentripplanner.

the class GraphPathFinder method createReversedTransitRequest.

private RoutingRequest createReversedTransitRequest(RoutingRequest originalReq, RoutingRequest options, Vertex fromVertex, Vertex toVertex, long arrDepTime, RemainingWeightHeuristic remainingWeightHeuristic) {
    RoutingRequest request = createReversedRequest(originalReq, options, fromVertex, toVertex, arrDepTime, new EuclideanRemainingWeightHeuristic());
    if ((originalReq.parkAndRide || originalReq.kissAndRide) && !originalReq.arriveBy) {
        request.parkAndRide = false;
        request.kissAndRide = false;
        request.modes.setCar(false);
    }
    request.maxWalkDistance = CLAMP_MAX_WALK;
    return request;
}
Also used : RoutingRequest(org.opentripplanner.routing.core.RoutingRequest) EuclideanRemainingWeightHeuristic(org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic)

Example 2 with EuclideanRemainingWeightHeuristic

use of org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic in project OpenTripPlanner by opentripplanner.

the class GraphPathFinder method getPaths.

/**
 * Repeatedly build shortest path trees, retaining the best path to the destination after each try.
 * For search N, all trips used in itineraries retained from trips 0..(N-1) are "banned" to create variety.
 * The goal direction heuristic is reused between tries, which means the later tries have more information to
 * work with (in the case of the more sophisticated bidirectional heuristic, which improves over time).
 */
public List<GraphPath> getPaths(RoutingRequest options) {
    RoutingRequest originalReq = options.clone();
    if (options == null) {
        LOG.error("PathService was passed a null routing request.");
        return null;
    }
    // Reuse one instance of AStar for all N requests, which are carried out sequentially
    AStar aStar = new AStar();
    if (options.rctx == null) {
        options.setRoutingContext(router.graph);
    // The special long-distance heuristic should be sufficient to constrain the search to the right area.
    }
    // If this Router has a GraphVisualizer attached to it, set it as a callback for the AStar search
    if (router.graphVisualizer != null) {
        aStar.setTraverseVisitor(router.graphVisualizer.traverseVisitor);
    // options.disableRemainingWeightHeuristic = true; // DEBUG
    }
    // Without transit, we'd just just return multiple copies of the same on-street itinerary.
    if (!options.modes.isTransit()) {
        options.numItineraries = 1;
    }
    // FORCING the dominance function to weight only
    options.dominanceFunction = new DominanceFunction.MinimumWeight();
    LOG.debug("rreq={}", options);
    // Choose an appropriate heuristic for goal direction.
    RemainingWeightHeuristic heuristic;
    RemainingWeightHeuristic reversedSearchHeuristic;
    if (options.disableRemainingWeightHeuristic) {
        heuristic = new TrivialRemainingWeightHeuristic();
        reversedSearchHeuristic = new TrivialRemainingWeightHeuristic();
    } else if (options.modes.isTransit()) {
        // Only use the BiDi heuristic for transit. It is not very useful for on-street modes.
        // heuristic = new InterleavedBidirectionalHeuristic(options.rctx.graph);
        // Use a simplistic heuristic until BiDi heuristic is improved, see #2153
        heuristic = new InterleavedBidirectionalHeuristic();
        reversedSearchHeuristic = new InterleavedBidirectionalHeuristic();
    } else {
        heuristic = new EuclideanRemainingWeightHeuristic();
        reversedSearchHeuristic = new EuclideanRemainingWeightHeuristic();
    }
    options.rctx.remainingWeightHeuristic = heuristic;
    /* In RoutingRequest, maxTransfers defaults to 2. Over long distances, we may see
         * itineraries with far more transfers. We do not expect transfer limiting to improve
         * search times on the LongDistancePathService, so we set it to the maximum we ever expect
         * to see. Because people may use either the traditional path services or the 
         * LongDistancePathService, we do not change the global default but override it here. */
    options.maxTransfers = 4;
    // Now we always use what used to be called longDistance mode. Non-longDistance mode is no longer supported.
    options.longDistance = true;
    /* In long distance mode, maxWalk has a different meaning than it used to.
         * It's the radius around the origin or destination within which you can walk on the streets.
         * If no value is provided, max walk defaults to the largest double-precision float.
         * This would cause long distance mode to do unbounded street searches and consider the whole graph walkable. */
    if (options.maxWalkDistance == Double.MAX_VALUE)
        options.maxWalkDistance = DEFAULT_MAX_WALK;
    if (options.maxWalkDistance > CLAMP_MAX_WALK)
        options.maxWalkDistance = CLAMP_MAX_WALK;
    long searchBeginTime = System.currentTimeMillis();
    LOG.debug("BEGIN SEARCH");
    List<GraphPath> paths = Lists.newArrayList();
    while (paths.size() < options.numItineraries) {
        // TODO pull all this timeout logic into a function near org.opentripplanner.util.DateUtils.absoluteTimeout()
        int timeoutIndex = paths.size();
        if (timeoutIndex >= router.timeouts.length) {
            timeoutIndex = router.timeouts.length - 1;
        }
        double timeout = searchBeginTime + (router.timeouts[timeoutIndex] * 1000);
        // Convert from absolute to relative time
        timeout -= System.currentTimeMillis();
        // Convert milliseconds to seconds
        timeout /= 1000;
        if (timeout <= 0) {
            // Catch the case where advancing to the next (lower) timeout value means the search is timed out
            // before it even begins. Passing a negative relative timeout in the SPT call would mean "no timeout".
            options.rctx.aborted = true;
            break;
        }
        // Don't dig through the SPT object, just ask the A star algorithm for the states that reached the target.
        aStar.getShortestPathTree(options, timeout);
        if (options.rctx.aborted) {
            // Search timed out or was gracefully aborted for some other reason.
            break;
        }
        List<GraphPath> newPaths = aStar.getPathsToTarget();
        if (newPaths.isEmpty()) {
            break;
        }
        // Do a full reversed search to compact the legs
        if (options.compactLegsByReversedSearch) {
            newPaths = compactLegsByReversedSearch(aStar, originalReq, options, newPaths, timeout, reversedSearchHeuristic);
        }
        // Find all trips used in this path and ban them for the remaining searches
        for (GraphPath path : newPaths) {
            // path.dump();
            List<AgencyAndId> tripIds = path.getTrips();
            for (AgencyAndId tripId : tripIds) {
                options.banTrip(tripId);
            }
            if (tripIds.isEmpty()) {
                // This path does not use transit (is entirely on-street). Do not repeatedly find the same one.
                options.onlyTransitTrips = true;
            }
        }
        paths.addAll(newPaths.stream().filter(path -> {
            double duration = options.useRequestedDateTimeInMaxHours ? options.arriveBy ? options.dateTime - path.getStartTime() : path.getEndTime() - options.dateTime : path.getDuration();
            return duration < options.maxHours * 60 * 60;
        }).collect(Collectors.toList()));
        LOG.debug("we have {} paths", paths.size());
    }
    LOG.debug("END SEARCH ({} msec)", System.currentTimeMillis() - searchBeginTime);
    Collections.sort(paths, new PathComparator(options.arriveBy));
    return paths;
}
Also used : RemainingWeightHeuristic(org.opentripplanner.routing.algorithm.strategies.RemainingWeightHeuristic) TrivialRemainingWeightHeuristic(org.opentripplanner.routing.algorithm.strategies.TrivialRemainingWeightHeuristic) EuclideanRemainingWeightHeuristic(org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic) AgencyAndId(org.onebusaway.gtfs.model.AgencyAndId) AStar(org.opentripplanner.routing.algorithm.AStar) GraphPath(org.opentripplanner.routing.spt.GraphPath) EuclideanRemainingWeightHeuristic(org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic) InterleavedBidirectionalHeuristic(org.opentripplanner.routing.algorithm.strategies.InterleavedBidirectionalHeuristic) TrivialRemainingWeightHeuristic(org.opentripplanner.routing.algorithm.strategies.TrivialRemainingWeightHeuristic) RoutingRequest(org.opentripplanner.routing.core.RoutingRequest) DominanceFunction(org.opentripplanner.routing.spt.DominanceFunction)

Example 3 with EuclideanRemainingWeightHeuristic

use of org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic in project OpenTripPlanner by opentripplanner.

the class GraphPathFinder method compactLegsByReversedSearch.

/**
 * Do a full reversed search to compact the legs of the path.
 *
 * By doing a reversed search we are looking for later departures that will still be in time for transfer
 * to the next trip, shortening the transfer wait time. Also considering other routes than the ones found
 * in the original search.
 *
 * For arrive-by searches, we are looking to shorten transfer wait time and rather arrive earlier.
 */
private List<GraphPath> compactLegsByReversedSearch(AStar aStar, RoutingRequest originalReq, RoutingRequest options, List<GraphPath> newPaths, double timeout, RemainingWeightHeuristic remainingWeightHeuristic) {
    List<GraphPath> reversedPaths = new ArrayList<>();
    for (GraphPath newPath : newPaths) {
        State targetAcceptedState = options.arriveBy ? newPath.states.getLast().reverse() : newPath.states.getLast();
        if (targetAcceptedState.stateData.getNumBooardings() < 2) {
            reversedPaths.add(newPath);
            continue;
        }
        final long arrDepTime = targetAcceptedState.getTimeSeconds();
        LOG.debug("Dep time: " + new Date(newPath.getStartTime() * 1000));
        LOG.debug("Arr time: " + new Date(newPath.getEndTime() * 1000));
        // find first/last transit stop
        Vertex transitStop = null;
        long transitStopTime = arrDepTime;
        while (transitStop == null) {
            if (targetAcceptedState.backEdge instanceof TransitBoardAlight) {
                if (options.arriveBy) {
                    transitStop = targetAcceptedState.backEdge.getFromVertex();
                } else {
                    transitStop = targetAcceptedState.backEdge.getToVertex();
                }
                transitStopTime = targetAcceptedState.getTimeSeconds();
            }
            targetAcceptedState = targetAcceptedState.getBackState();
        }
        // find the path from transitStop to origin/destination
        Vertex fromVertex = options.arriveBy ? options.rctx.fromVertex : transitStop;
        Vertex toVertex = options.arriveBy ? transitStop : options.rctx.toVertex;
        RoutingRequest reversedTransitRequest = createReversedTransitRequest(originalReq, options, fromVertex, toVertex, arrDepTime, new EuclideanRemainingWeightHeuristic());
        aStar.getShortestPathTree(reversedTransitRequest, timeout);
        List<GraphPath> pathsToTarget = aStar.getPathsToTarget();
        if (pathsToTarget.isEmpty()) {
            reversedPaths.add(newPath);
            continue;
        }
        GraphPath walkPath = pathsToTarget.get(0);
        // do the reversed search to/from transitStop
        Vertex fromTransVertex = options.arriveBy ? transitStop : options.rctx.fromVertex;
        Vertex toTransVertex = options.arriveBy ? options.rctx.toVertex : transitStop;
        RoutingRequest reversedMainRequest = createReversedMainRequest(originalReq, options, fromTransVertex, toTransVertex, transitStopTime, remainingWeightHeuristic);
        aStar.getShortestPathTree(reversedMainRequest, timeout);
        List<GraphPath> newRevPaths = aStar.getPathsToTarget();
        if (newRevPaths.isEmpty()) {
            reversedPaths.add(newPath);
        } else {
            List<GraphPath> joinedPaths = new ArrayList<>();
            for (GraphPath newRevPath : newRevPaths) {
                LOG.debug("REV Dep time: " + new Date(newRevPath.getStartTime() * 1000));
                LOG.debug("REV Arr time: " + new Date(newRevPath.getEndTime() * 1000));
                List<GraphPath> concatenatedPaths = Arrays.asList(newRevPath, walkPath);
                if (options.arriveBy) {
                    Collections.reverse(concatenatedPaths);
                }
                GraphPath joinedPath = joinPaths(concatenatedPaths);
                if ((!options.arriveBy && joinedPath.states.getFirst().getTimeInMillis() > options.dateTime * 1000) || (options.arriveBy && joinedPath.states.getLast().getTimeInMillis() < options.dateTime * 1000)) {
                    joinedPaths.add(joinedPath);
                    if (newPaths.size() > 1) {
                        for (AgencyAndId tripId : joinedPath.getTrips()) {
                            options.banTrip(tripId);
                        }
                    }
                }
            }
            reversedPaths.addAll(joinedPaths);
        }
    }
    return reversedPaths.isEmpty() ? newPaths : reversedPaths;
}
Also used : Vertex(org.opentripplanner.routing.graph.Vertex) AgencyAndId(org.onebusaway.gtfs.model.AgencyAndId) TransitBoardAlight(org.opentripplanner.routing.edgetype.TransitBoardAlight) State(org.opentripplanner.routing.core.State) GraphPath(org.opentripplanner.routing.spt.GraphPath) RoutingRequest(org.opentripplanner.routing.core.RoutingRequest) EuclideanRemainingWeightHeuristic(org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic)

Aggregations

EuclideanRemainingWeightHeuristic (org.opentripplanner.routing.algorithm.strategies.EuclideanRemainingWeightHeuristic)3 RoutingRequest (org.opentripplanner.routing.core.RoutingRequest)3 AgencyAndId (org.onebusaway.gtfs.model.AgencyAndId)2 GraphPath (org.opentripplanner.routing.spt.GraphPath)2 AStar (org.opentripplanner.routing.algorithm.AStar)1 InterleavedBidirectionalHeuristic (org.opentripplanner.routing.algorithm.strategies.InterleavedBidirectionalHeuristic)1 RemainingWeightHeuristic (org.opentripplanner.routing.algorithm.strategies.RemainingWeightHeuristic)1 TrivialRemainingWeightHeuristic (org.opentripplanner.routing.algorithm.strategies.TrivialRemainingWeightHeuristic)1 State (org.opentripplanner.routing.core.State)1 TransitBoardAlight (org.opentripplanner.routing.edgetype.TransitBoardAlight)1 Vertex (org.opentripplanner.routing.graph.Vertex)1 DominanceFunction (org.opentripplanner.routing.spt.DominanceFunction)1