Uros Stankovic created SPARK-60421:
--------------------------------------
Summary: Fix behaviour of vector avg function for special float
values
Key: SPARK-60421
URL: https://issues.apache.org/jira/browse/SPARK-60421
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 4.4.0
Reporter: Uros Stankovic
The existing update formula of vector_avg implementation can overflow its
intermediate subtraction or turn an infinite average into NaN after a finite
input. This makes updates disagree with merging singleton buffers and makes the
result depend on input order.
The following cases show two successive inputs for one vector element:
| Input sequence | Current behaviour | Correct behaviour |
|---|---|---|
| `Float.MaxValue`, then `-Float.MaxValue` | `-Infinity` | `0.0f` |
| `-Float.MaxValue`, then `Float.MaxValue` | `Infinity` | `0.0f` |
| `Infinity`, then `Infinity` | `NaN` | `Infinity` |
| `-Infinity`, then `-Infinity` | `NaN` | `-Infinity` |
| `Infinity`, then `-Infinity` (either order) | `NaN` | `NaN` |
| `Infinity`, then a finite value | `NaN` | `Infinity` |
| `-Infinity`, then a finite value | `NaN` | `-Infinity` |
| A finite value, then `Infinity` | `Infinity` | `Infinity` |
| A finite value, then `-Infinity` | `-Infinity` | `-Infinity` |
| `NaN` and any value (either order) | `NaN` | `NaN` |
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
