Update variance/stddev to work with single values #4847
Conversation
Following Postgres: - var/stddev of single element is NULL - var_pop/stddev_pop of single element is 0
github-actions
bot
added
core
| Ok(ScalarValue::Float64(match self.count { | ||
| 0 => None, | ||
| 1 => { | ||
| if matches!(self.stats_type, StatsType::Population) { |
There was a problem hiding this comment.
Nit: In this case I think if let would be slightly more idiomatic.
|
Hmm, some tests seem to be failing -- needs some investigation |
|
Ok, the tests are sorted out now. Thanks for taking a look! |
|
@ozankabak -- thank you as well for your help reviewing. It is really appreciated |
|
Benchmark runs are scheduled for baseline = dcd52ee and contender = 3d75bb8. 3d75bb8 is a master commit associated with this PR. Results will be available as each benchmark for each run completes. |
* Update variance/stddev to work with single values Following Postgres: - var/stddev of single element is NULL - var_pop/stddev_pop of single element is 0 * Fix tests * matches! to if let * fix test_stddev_1_input test (cherry picked from commit 3d75bb8)
* Update variance/stddev to work with single values Following Postgres: - var/stddev of single element is NULL - var_pop/stddev_pop of single element is 0 * Fix tests * matches! to if let * fix test_stddev_1_input test (cherry picked from commit 3d75bb8)
Which issue does this PR close?
Closes #4843.
Rationale for this change
Rather than crash when performing a variance-based aggregation on a single value, follow Postgres' semantics:
var/stddevof single element isNULLvar_pop/stddev_popof single element is 0There's documentation of this behavior in Snowflake (https://docs.snowflake.com/en/sql-reference/functions/stddev.html), and it's what Postgres does, but in searching briefly I haven't found this behavior described in the Postgres docs.
Are these changes tested?
I added a pair of sqllogictest tests for the one and two element cases.
Are there any user-facing changes?
Queries that used to crash when aggregating single values with variance-based aggregations will no longer crash.