Better ways of handling distribution shift?

[emilycantrell]

Currently we have an indicator for whether the data is time-shifted, as a simple attempt to adjust for distribution shift. There is literature on strategies for handling distribution shift. I'll investigate whether there is another simple strategy we can try, e.g., perhaps something from this list https://out-of-distribution-generalization.com/