Closed
Description
Found in version
- 6.3.0-bc11
Steps to reproduce
- Create a multi metric ML job (e.g. for the farequote dataset with 15m buckets and
mean(responsetime) partition_field_name=airline
) - Create a
28d
forecast on the results
Actual result
The forecast shows a trend to infinity or zero in an exponential manner, which is sometimes unexpected when looking at the actual data the forecast is based on. And it is a significant change compared to the results of the same forecast in version 6.2.5:
Additional information
In other datasets with more base data for the forecast (e.g. cloudwatch
) the same trend can be seen: