Open jexp opened 5 years ago
Hi min, max and sum is more like an old relict. Those values are helpfull but they need a full iteration after the algo terminates. Having more statistics or a histogram (as you mentioned in #652 ) would make more sense in my opinion. But then we should focus on implementing a robust toolkit for this task which can be used in other algos as well. Im more concered about Aragorn and the other 0.0's. Are those nodes disconnected?
we started to add min, max, mean, stddev, p25, p50,p75,p90,p95,p99,p999,p100
for Jaccard and I think we should add it to all the other non-streaming algorithms too.
Agree on the toolkit. We used (Atomic)HdrHistogram to gather those stats.
Just a comment: I think min and max are useful to normalize the centrality, or that could come as an optional param too.
It should not be -1 on all accounts.