rstudio-conf-2020 / intro-to-ml-tidy Goto Github PK

uses ames instead of ames_train

cc @JosiahParry

• policies
• CoC
• Intros
• WiFi
• RSP setup instructions

Right before "Your turn 2" in the 01-prediction slides (https://conf20-intro-ml.netlify.com/materials/01-predicting/)

You use the custom fit_dat function to fit the lm model

lm_spec <- 
   linear_reg() %>% # Pick linear regression
   set_engine(engine = "lm") # set engine

fit_data(Sale_Price ~ Gr_Liv_Area, model = lm_spec, data = ames)

You can also do it straight in parsnip now without using the custom fit_data function now

lm_spec <- parsnip::linear_reg() %>%
  parsnip::set_engine(engine = "lm")

parsnip::fit(lm_spec, Sale_Price ~ Gr_Liv_Area, data = ames)

It's even pipe-able!

Not sure if this is a new feature that was added or just to stay consistent with the trees/random forest slides later on.

option(scipen=17)

@apreshill,

When passing a workflow object to tune::collect_metrics() the resulting error message is Error: All of the models failed. (see attached example)

A more informative error message would remind the user to pass a fit_split object instead.

Thanks!

In the 06-recipes slides:

I think this should've read ames_training instead of just ames.
There are a lot of slides before it that use ames where it should be using ames_training, but this example gets built on to train the model, so it seems especially important here.

#18 is also related

This is one of my favorite blog posts that talk about the various regression metrics.

https://www.h2o.ai/blog/regression-metrics-guide/