forecast.mbl() gains a new argument names_to = ".model"?

[earowang]

hence we can easily rename the .model column.

[mitchelloharawild]

Sounds reasonable, although ".model" is assumed in a variety of places (plotting, accuracy, etc.). It will probably need to be stored as an attribute of the fbl_ts

[earowang]

isn't it part of key?

[mitchelloharawild]

Yes, but ".model" is treated differently from key in some places. For example when plotting, the model is used for the colour, other keys are row facets, and response variable (if multivariate) are column facets.

[earowang]

can setdiff() work out the difference?

[mitchelloharawild]

Difference from what? The original tsibble keys aren't stored, just the current ones (inclusive of .model).
Perhaps in the plot it could be used as the data may be available, although it is possible to plot the forecasts without data.

[mitchelloharawild]

.model is also used with augment(), glance(), tidy(), etc.
Should all these functions gain this argument?

[earowang]

I think so, whenever to "longer" format.

Can we just have model instead of mdl_nm?

[mitchelloharawild]

Consolidated into #192

[mitchelloharawild]

Reopening as I don't have time to include this in today's release, which closes #192.