Closed iamyihwa closed 8 months ago
Hey @iamyihwa. As described here, it's very important that your dataframe is partitioned by series for the distributed processing to work correctly, so you have to repartition your dataframe or provide a value for num_partitions
to the DistributedMLForecast
constructor.
Thanks a lot @jmoralez ! By using .repartitionByRange as suggested, the problem is solved!
What happened + What you expected to happen
Hi, I 've witnessed a very strange case, where the timesteps of resulting forecasts don't correspond to the number of horizon given.
There are a few issues. (1) The resulting time series don't correspond to the number of horizons given (13 ).
(3) Also they are not consistent. In this example, where 3 different unique_ids are provided, 2 of them have different number of time steps compared to the other one. (See figure 1)
Versions / Dependencies
0.10.0
Reproduction script
Issue Severity
None