Another aggregate function we’ll need to look at is median().

That should in theory map to the existing quantile kernel and the hopefully-coming-soon hash_quantile kernel with q = 0.5, but that will not work right now for several reasons (see https://issues.apache.org/jira/browse/ARROW-13309).

However we also have a pair of
approximate quantile kernels: tdigest and hash_tdigest that we could map median() to.

Reporter: Nicola Crane / @thisisnic
Assignee: Ian Cook / @ianmcook

Related issues:

• [R] Use Arrow engine for summarize() by default (blocks)
• [C++] Kernel to return Array of elements at index of list in ListArray (is blocked by)
• [C++] tdigest, quantile return empty arrays when nulls not skipped (is blocked by)
• [C++] Add appx_median, hash_appx_median functions (is blocked by)
• [R] summarize() should not eagerly evaluate (is blocked by)
• [R] Bind median() and quantile() to exact not approximate median and quantile (is related to)
• [C++] Implement hash_aggregate approximate quantile kernel (depends upon)

PRs and other links:

• GitHub Pull Request #11018

_{Note: This issue was originally created as ARROW-13772. Please see the migration documentation for further details.}