Investigate alternatives for pivotting dataframes

[Peter9192]

In our workflows, the target is always unique for each location/year. However, sometimes the predictors are timeseries for the same point/year. We need a way to represent time series as input features.

Currently, our approach is to pivot the dataframe, such that each day/month/entry in the timeseries is one column. This leads to many columns with complex names, such as

year	geometry	tmax|1	tmax|2	tmax|3	...	tmin|9	tmin|10	tmin|11	tmin|12

Alternatively, we could

• Store time series as such in a single pandas cell
• Switch to xarray or something similar to better represent multidimensional data
• ...?