which package to use, gluonts, darts, ...

[emptymalei]

Developing a well-engineered model takes a long time and a lot of engineering/boring efforts. In light of this, we choose a package to deal with the tedious engineering tasks.

There are some popular packages on the market already.

1. GluonTS
   1. Can deal with large datasets as they use iterstyle datasets.
   2. Iterstyle datasets usually take more engineering work on GPU, however, we probably won't work with GPUs anyways.
   3. Good proprocessing tools but less efforts in development into backtesting.
   4. Very steep learning curve.
   5. We have some connections to gluonts in the group. For example, there is already an implementation of the model here.
2. Darts
   1. A nice TimeSeries object for all kinds of time series datasets.
   2. Not easy to deal with datasets that don't fit into memory. But we probably won't have this problem.
   3. Good backtesting tools.
   4. Easy to get started.
   5. No connections to the developers.