how to deal with signals with partially missing data?

[dsweber2]

This may be something we can handle in extend_ahead and family, or it may have to be an additional input

[dshemetov]

Options

• Catch these errors and halt
• Catch error and forecast nothing
• Catch error and impute gaps in data

[nmdefries]

Broadly, find a way to impute missing data.

epipredict fns complain about NAs a lot. Want at least one fn that can handle these. Maybe an arg can set the requested behavior (what we've done in other scripts: 0-fill, mean-fill, smoothing between two closest points)

[dshemetov]

Related: cmu-delphi/epipredict#106

[dshemetov]

Related #59

[dshemetov]

closed by #55

handled by using drop_na on the input data to the pipeline. we can open another issue if find we need imputation.