Joshua Andersen

Results 6 comments of Joshua Andersen

These features seem fine though I will need to more research on implementations for computing intersection cardinality.

@kuk HyperLogLogs do not store the set, which is why they are memory efficient, so you cannot compute expressions like `cardinality(A ∪ B)` or `cardinality(A ∩ B)` with the data...

Yes but this isn't done using a union operation. Instead there is a register by register comparison and the maximum value is the winner. Then the cardinality estimation of the...

I should add that because you take the maximum value estimating the cardinality of a union is straightforward since the algorithm only needs the maximum values. However for intersections the...

And what happens when each estimate is 2% off in the same direction? The total error is then approximately 6% which is 3 times the error you normally get. Now...

Thanks for reporting this. I suspect this is related to serialization/deserialization of the registers when in sparse representation. I will investigate. As a temporary fix, you can do `sparse=False` in...