[
https://issues.apache.org/jira/browse/BEAM-7525?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17188061#comment-17188061
]
Monica Song edited comment on BEAM-7525 at 9/1/20, 12:32 AM:
-------------------------------------------------------------
allowing a customizable hash function would also mean having the user provide
the max hash size and sample size space (since hashes can be signed/unsigned
and 32/64/128 bit), as well as an additional check to make sure the hash
function returns an integer fingerprint (not a byte string). This may be a knob
that relies too much on knowledge of hash functions.
was (Author: monicadsong):
allowing a customizable hash function would also mean having the user provide
the max hash size and sample size space (since hashes can be signed/unsigned
and 32/64/128 bit). This may be a knob that relies too much on knowledge of
hash functions.
> support hash_fn for Python ApproximateUniqueCombineFn transform
> ---------------------------------------------------------------
>
> Key: BEAM-7525
> URL: https://issues.apache.org/jira/browse/BEAM-7525
> Project: Beam
> Issue Type: Task
> Components: sdk-py-core
> Reporter: Hannah Jiang
> Priority: P3
>
> ApproximateUniqueCombineFn is using default hash function for estimation.
> We can pass a hash_fn and overwrite default hash function to support better
> estimation performance.
> Parent issue: BEAM-6693
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
