Hi there
We have some questions about A* Implementation; we were thinking to use the
TraversalAStar, since we really like the traversal framework and it's
simpler to wotk on it
By the way, also by using the not traversal AStar (the classical
implementatin), as estimateevaluator we used
the CommonEvaluators.geoEstimateEvaluator; now in our routing calculation,
we noticed some strange paths (we checked for several points and we
compared the result with google and other routing systems)
We, then, created a custom estimateevaluator; we simply created the
following:
EstimateEvaluator<Double> estimateEvaluator = new EstimateEvaluator<Double
>()
{
public Double getCost( final Node node, final Node goal )
{
double dx = (Double)
node.getProperty(OSMAttribute.LONGITUDE_PROPERTY
) - (Double) goal.getProperty( OSMAttribute.LONGITUDE_PROPERTY );
double dy = (Double) node.getProperty(
OSMAttribute.LATITUDE_PROPERTY
) - (Double) goal.getProperty( OSMAttribute.LATITUDE_PROPERTY );
double result = Math.sqrt( Math.pow( dx, 2 ) + Math.pow( dy,
2 ) );
return result;
}
};
Well by using this evaluator the returned path are pretty similar to the
ones returned by google and the other softwares. Does anybody know the
reason of this behaviour? Are we missing anything?
Moreover.....may you tell us the maturity of the TraversalAStar
implementation? In the source code we saw that it's in experimental status
but for the few tests we did it seems to work good....is it possible to use
it in production environment?
Thank you
Angelo
Il giorno lunedì 19 maggio 2014 09:04:51 UTC+2, Angelo Immediata ha scritto:
>
> Hi there
>
> With my colleague, we are are buillding a route system by using neo4j
> 2.0.3; so we are suing A* and Dijkstra algorithms in order to calculate the
> shortest path,
> I was wondering if the relationships number can affect the algorithm
> perfomance. I mean, we have a graph with around 1 million (or more) of
> nodes and 50 million of relationships. We have several types of
> relationship; specifically we have:
>
> - relationships for cars: the most of relationships are of this type
> - relationships for bikes
> - relationships for pedestrian
> - relationships for public transports
>
> When we execute Dijkstra and/or A* we can specify, in our PathExpander,
> the type of the relationships we want to consider during the traverser, so,
> my sensation is that the relationships number should not affect algorithm
> performance since we will sparsely (almost never) consider all the
> relationships types. Am I right?
>
> Thak you
> Angelo
>
--
You received this message because you are subscribed to the Google Groups
"Neo4j" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
