Github user fmcquillan99 commented on the issue:
https://github.com/apache/madlib/pull/256
Oh I see, with the averaging approach:
buffer_size = avg_num_rows_per_segment / num_segments
= 21.5 / 2
= 10.75
and rounding up we get 11.
Can you think of any drawbacks of using this approach?---
