The X_df dataframe is sorted by id and time and was then sliced in each step by moving an n_series sized window, which was wrong. The correct slice is taking a range that starts at the current predict step and offsets each element by h, thus taking the first date of each id in step 0, the second date of each id in step 1 and so on.
Checklist:
[x] This PR has a meaningful title and a clear description.
Description
The
X_df
dataframe is sorted by id and time and was then sliced in each step by moving ann_series
sized window, which was wrong. The correct slice is taking a range that starts at the current predict step and offsets each element byh
, thus taking the first date of each id in step 0, the second date of each id in step 1 and so on.Checklist: