Question on filling values and results save
sami-ka opened this issue · comments
Hi,
Thanks a lot for the great work you put into this package!
I had to implement time series features on my own before discovering this package, and I have 2 questions about what is doable with functime:
- Let's say you want to compute the rolling average number of purchase by day in the last 3 months for all possible pairs of (customer,product) for each day, out of a transaction dataset. There might be some days where there is no purchase for some pairs (customer, product), in that case you want to impute a 0 for any day with missing purchase.
If I want to compute the average margin percentage each day, it will be different as I would like to leave the days without purchase and drop the NaNs.
Is this customized imputation by value to aggregate feasible right now? Or do you expect to have the whole set of dates and (customer, product) pairs given with the already filled nan values? Because this would imply an extra step compared to working directly with the transaction dataset. - Is there a way to save the time series feature computation for all possible dates and pairs? To keep going with my previous example, I would like to keep the average number of purchase by day for every possible dates and store it somewhere, instead of having the feature computed only for pairs and dates in my transaction dataset. This would allow me to not recompute everything when I want feature information at a specific date that is not present in my transaction dataset.
Looking forward to your answers!
