Hello, I have been using HierarchicalForecast Meth...
# generalf
Faruk Erdoğan Buldur
09/11/2024, 1:02 PMHello, I have been using HierarchicalForecast Methods on M5 Dataset, but when I try to reconcile store level predictions with following hierarchy it causes Kernel to crash. I have tried to find what might cause the problem but I have not able to spot the problem. Any idea from your side ?
• Total
• Total/Item
• Total/Item/Store
m
Marco
09/11/2024, 1:25 PM@Olivier?
o
Olivier
09/11/2024, 1:57 PMCould you share the code that you run? (a piece of code that I can also run, producing the error)
h
Hakan Ateşli
09/12/2024, 7:08 AMHello, you can find our work in the notebook below. I am getting an prediction for the Item level with MLForecast. I am getting a Naive prediction for the State and Store levels. Then I am trying to apply reconciliation with TopDown at the State and Store levels using the Item level. The kernel is not crashing for now. But the result is not returning. Despite working for 2-3 hours, the result was not returned even for a single fold.
o
Olivier
09/12/2024, 7:34 AMThanks, I'm running the notebook right now
Olivier
09/12/2024, 7:44 AMOn my machine I can run everything fine, but the reconciliation is insanely slow. I stopped it after 10 minutes. So I used
MiddleOutSparseMiddleOut😁 1
Olivier
09/12/2024, 7:45 AMAnyways, hope this helps, let me know if the swap makes it work for you.
We haven't yet updated the docs for this as Sparse methods have only recently been added (and are stillm being added)
f
Faruk Erdoğan Buldur
09/12/2024, 7:46 AMWe have tried sparse Summing Matrix but that was not helpful in terms of speed and we were unaware of
MiddleOutSparse👍 1
o
Olivier
09/12/2024, 7:46 AMNote that you need to update hierarchicalforecast to the last version
👍 2
h
Hakan Ateşli
09/12/2024, 6:27 PMIt worked. Thanks for the support.