search

cmu-delphi / flu-hosp-forecast

1 stars 0 forks source link

Compare uses a shorter training window to a longer training window #40

Closed dshemetov closed 9 months ago

dshemetov commented 11 months ago

Our "no window" training models currently use 20 years of training data. This is likely overkill (and possibly not relevant to today's data). Try using 2 years of data (after 2021 to match CDC baseline).

dshemetov commented 9 months ago

The "no window" training models actually use about 4 years of data. See #39 for more details.

Closing this specific issue. Training windows might be worth exploring another time.