Citation Context

..., ALT works well in dynamic scenarios [15]. Hence, we expected that CALT (CoreALT) also works well in dynamic and time-dependent scenarios, which recently has been shown in [10]. Second, we showed in =-=[6]-=- that pure ALT is a very robust technique with respect to the input. Finally, ALT suffers from the critical drawback of high memory consumption—we have to store two distances per node and landmark—whi...

Citation Context

...ocessing-based techniques for computing point-to-point shortest paths have been very successful on road networks [8, 16], but their adaption to public transit networks [2, 10] is harder than expected =-=[1, 3, 4]-=-. The problem of computing “best” journeys comes in several variants [14]: The simplest, called earliest arrival, takes a departure time as input, and determines a journey that arrives at the destinat...

Citation Context

...bels L that are dominated by a label from the target vertex’s bag. We do not use goaldirection for MLC; although helpful for long-distance rail networks [15], it does not help on dense urban networks =-=[4]-=-. Even with these improvements, MLC is much more costly than plain Time-Dijsktra: not only must it scan the same vertex multiple times, but it also has to handle more complicated data structures, such...

Citation Context

...odel has been introduced in [22]. The model has been generalized in [19] in order to deal with realistic transfers. Since then, this realistic model has been used for many experimental studies, e.g., =-=[15, 20, 2]-=-; most of them focusing on faster speed-up techniques or multi-criteria optimization for timetable information. However, [22] enriched the simple time-expanded graph by shortcuts and [20] introduced m...

Citation Context

...all number of memory accesses that are necessary. We observe speedups over Dijkstra’s algorithm of 75–140. This is about an order of magnitude more than what we would expect from previous work, e.g. [=-=Bauer, Delling, and Wagner, 2010-=-; Delling, 2009] report factors of 7.0–8.5 when evaluating the performance of uniALT on a long distance timetable network. While it is not surprising to see a speedup of A∗ over Dijkstra’s algorithm, ...

Citation Context

... e-mail: dieter.rautenbach@tu-ilmenau.de 1in the artificial-intelligence setting [11, 16] and were later discussed as exact algorithms in the general context and in combination with other techniques =-=[4, 5, 15, 23, 32]-=-. There has been tremendous experimental effort to evaluate combinations of these techniques [9, 21]. See [41] for a comprehensive overview on speed-up techniques published during the last years. The ...

Citation Context

...ion of recorded timetables where the journey was successful and allowed to arrive at the destination on time. We provide efficient algorithms for computation of this measure. 2 Related Work The problem of finding a fastest journey (according to the planned timetable) using public transportation has been extensively studied in the literature. Common approaches model the transportation network as a graph and compute a shortest path in this graph (see [12] for a survey). Various improvements have been developed, and experimental studies suggest that these can also be used in practice (see, e.g., [2, 5, 14]). Recent approaches avoid the construction of a graph and process the timetable directly [6, 7]. For example, Delling et al. [6] describe an approach which is centered around transportation lines (such as train or bus lines) and which can be used to find all pareto-optimal journeys when the arrival time and the number of stops are considered as criteria. Bast et al. [1] observe that for two given stops, we can find and encode each sequence of intermediate transfer stations (i.e., stations where we change from one line to another) that can lead to an optimal route. The set of these sequences o...