Things to improve for next year

[ShirleyBWang]

EDA:

• Don't use featurePlot(); just stick with ggplot
• Be clearer about different plots for categorical vs continuous labels
• Use skimr instead of summaryTools (lots of errors in installing/loading summaryTools)

[jmgirard]

Conceptual Introduction:

• Spend a little more time explaining how cross-validation works

[jmgirard]

Consider just using CV as the final test set performance next time instead of nesting CV inside of holdout CV

Instead of baking test set by hand and doing performance manually, use broom::augment to add predictions to test set (this is much easier and less confusing, but requires parsnip)

[jmgirard]

Try to shorten and streamline tidyverse primer. Felt rushed and may not have helped the people who most need it (while boring the people who don't need it).

[jmgirard]

For 2C, slides 13 and 14 have some extra information about arguments that aren't necessary and it may be more streamlined (and less confusing) to remove this.

Also, consider redesigning 2C to maybe occur after/during the GLMNET section. Without the context of tuning to justify the inner CV, it is confusing.

[jmgirard]

For 2D, slide 5 there are some typos: should be recipe(surv_train, ...) and probably read_csv().

[jmgirard]

For 2D, slide 5 there are some typos: should be recipe(surv_train, ...) and probably read_csv().

Fixed in https://github.com/PittMethods/appliedml/commit/68de1d8ad7a37aad0f8d9d12efd529bb35edc5c1