Github user hvanhovell commented on the pull request:
https://github.com/apache/spark/pull/8362#issuecomment-133566608
Thanks.
I was aiming for compatibility with the existing approxCountDistinct, but
we can also implement HLL++. HLL++ introduces three (orthogonal) refinements:
64-bit hashing, better low cardinality corrections and a sparse encoding
scheme. The first two refinements are easy to add. The third will require a bit
more effort.
Unit testing this is a bit of a challenge. End-to-end (blackbox) testing is
no problem, as long as we know what the result should be, or if we do random
testing (results should be within 5% of the actual value). Testing parts of the
algorithm is a bit of a PITA:
* It is hard to reason about the results (the updated registers) HLL
produces.
* Register access code and HLL code are intertwined.
Both the
[ClearSpring](https://github.com/addthis/stream-lib/blob/master/src/main/java/com/clearspring/analytics/stream/cardinality/HyperLogLog.java)
and
[AggregateKnowledge](https://github.com/aggregateknowledge/java-hll/blob/master/src/main/java/net/agkn/hll/HLL.java)
implementations resort to blackbox testing. I will create some blackbox tests.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
