Our implementation of the AtmicDoubleArray is rather crude and i think noone cares about the scaleFactor. And in reality its impossible to calculate a good value since this depends on the algorithm, the node count, weights etc. It seems we often run into issues with our centrality-algos because of that. Anyway.. im thinking of changing the format to IEEE 754 by using javas builtin methods Double.toLongBits(..) etc. Then we can get rid of the optional configuration parameter scaleFactor. If we also implement a Float based 32bit version the user can add a config param for high or low precision. The Float Version just has to cast to double.
Our implementation of the AtmicDoubleArray is rather crude and i think noone cares about the
scaleFactor
. And in reality its impossible to calculate a good value since this depends on the algorithm, the node count, weights etc. It seems we often run into issues with our centrality-algos because of that. Anyway.. im thinking of changing the format to IEEE 754 by using javas builtin methodsDouble.toLongBits(..)
etc. Then we can get rid of the optional configuration parameterscaleFactor
. If we also implement a Float based 32bit version the user can add a config param for high or low precision. The Float Version just has to cast to double.