Redesign ML units
Opened this issue · 6 comments
bensoltoff commented
Consider Alison Hill's revised tidymodels
workshop. Still pretty similar to the old one, but has some distinct adjustments. Worth incorporating into a three-day unit on ML?
bensoltoff commented
Day 1
- Keep what I have
- Incorporate definition of stratified sampling
- Introduce
yardstick
for assessing model performance - Define ROC AUC
Day 2
- Keep as is
Day 3
- Introduce tree-based inference
- Define decision tree and hyperparameters
- Fit decision tree and random forest
- Tune hyperparameters for RF model using tuning grid
- Identify the best model and finalize the workflow
bensoltoff commented
Homework
- Explicitly incorporate tuning for second dataset
- Replace classification problem with more culturally-sensitive problem
bensoltoff commented
In-class exercises
Ensure they match the slides.
- Day 1
- Day 2
- Day 3
bensoltoff commented
Still need to revise hw07
bensoltoff commented
- Fix exercise files to be qmd not rmd
bensoltoff commented
- Fix
roc_curve()
slide. Should read
roc_curve(data, truth, ...)
truth = actual outcome
... = probability of one of the outcome classes