updates for upcoming parsnip

business-science / modeltime

Modeltime unlocks time series forecast models and machine learning in one framework

https://business-science.github.io/modeltime/

Other

522 stars 79 forks source link

updates for upcoming parsnip #8

Closed topepo closed 4 years ago

topepo commented 4 years ago

I'll be sending parsnip 0.1.2 to CRAN very soon and it has some differences in how to choose encodings from how modeltime currently does it.

Using arima_reg as an example, the model definition should use the new set_encoding() interface (rather than passing indicators directly). To make sure that parsnip (and, soon workflows) does no modifications to the predictors columns, use

set_encoding(
  model = "arima_reg",
  eng = "auto_arima",
  mode = "regression",
  options = list(
    predictor_indicators = "none",
    compute_intercept = FALSE,
    remove_intercept = FALSE
  )
)

They same type of declaration is required for each engine/model combination.

The current GH version of parsnip can be used for testing.

mdancho84 commented 4 years ago

Thanks Max. I'm on it. I'll work on the updates.

mdancho84 commented 4 years ago

@topepo I've updated. 2 minor issues in case you get this from others:

mars() models handle dates and features with model.matrix.earth. In case anyone gets the error, just tell them to use the workflows interface. Reference tidymodels/parsnip#341.
__Knitting with set_encodings():__ For some reason, I couldn't get my vignette, extending_model_time.Rmd to knit. It worked fine when I run code chunks locally, but knitting caused an error that I couldn't fix - training and predictions were causing an error that the date column was not present. My bandaid solution was to save the intermediate steps as .rds files and read them in the appropriate spots.

topepo commented 4 years ago

New parsnip was accepted to CRAN

mdancho84 commented 4 years ago

Nice work. Will send modeltime 0.0.2 to CRAN shortly.

mdancho84 commented 4 years ago

Modeltime 0.0.2 is on CRAN, which incorporates parsnip 0.1.2. We are good to go.