Before using the bidirectional Dijkstra with the edge-based traversal you need to create the shortcuts using edge-based local dijkstra searches (DijkstraOneToMany) and make it possible to store 'loops' for certain nodes.
> LevelEdgeFIlter don't give opportunity to encounter both directions in edge high-level node. why would you need this? Peter On 17.03.2015 19:36, dos65 wrote: > Forget my problem description. I found that "ch-algorithms" doesn't > work correctly in edge-based traversal mode because levels designed > only for nodes and LevelEdgeFIlter don't give opportunity to encounter > both directions in edge high-level node. > Should i introduce levels for edges or something else? > > > On 17.03.2015 13:57, Peter wrote: >> Hi! >> >>> I'm as yet in progress to full understanding how CH works. Described >>> problem i temporarily fixed by not to call "disconnect" after >>> contracting node (i don't know how important it is - in 0.3 was >>> "removeHigher2LowerEdges" flag). >> Not sure what you mean here, but this is no longer necessary in 0.4 >> >>> Unfortunately we can't afford be yours sponsor, but idea about crowd >>> funding sounds good. >>> Anyway i continue to trying to implement this. >> Trust me: implementing this on your own will be a lot more expensive >> than sponsoring :) >> >> Regards, >> Peter >> >>> 17.03.2015 13:16, Peter пишет: >>>> Hi, >>>> >>>>> it's not possible because both directions stop on high-level nodes on >>>> different edges. >>>> >>>> The search will be edge based and so the edges will be identical. >>>> >>>> There is an issue for this #270, where we list the major problems with >>>> turn costs and CH. This requires knowledge with CH and also some GH >>>> internals. I guess this will take us roughly 2-4 full time weeks. As >>>> this is relative unattractive long but very important (for us too) >>>> what >>>> do you think about a crowd funding campaign? >>>> Where every company in the need could sponsor us with only a >>>> fraction of >>>> the costs. (And we would also sponsor the majority of our time for >>>> this) >>>> >>>> Do not hesitate to contact me off-list if you or others are >>>> interested! >>>> >>>> Kind Regards, >>>> Peter >>>> >>>> On 16.03.2015 12:57, dos65 wrote: >>>>> Hi, I’m trying to implement turn costs for CH. >>>>> TurnCosts works only with edge-based traversal mode and the condition >>>>> for finding path is reaching same node and same edge for both >>>>> directions. But with levelGraph it's not possible because both >>>>> directions stop on high-level nodes on different edges. Do you have >>>>> any ideas about that? >> _______________________________________________ GraphHopper mailing list [email protected] https://lists.openstreetmap.org/listinfo/graphhopper
