Evaluating ML Performance, Resampling, and Workflows in "tidymodels" | R Tutorial (2021)
Vložit
- čas přidán 1. 04. 2021
- Subscribe to RichardOnData here: / @richardondata
Patreon: / richardondata
GitHub: github.com/RichardOnData/YouT...
Caret tutorial series:
Part 1: • Preprocessing Data in ...
Part 2: • Feature Elimination an...
Part 3: • Training and Tuning ML...
Part 4: • Creating ROC curves an...
Tidymodels:
Part 1: • Intro to machine learn...
In the previous tutorial series, we walked through the "caret" package in R for machine learning. We used the raw "GermanCredit" dataset, performed a brief exploration of it, and used the package to walk through a variety of steps: pre-processing, removing low information features, tuning hyperparameters, correcting for class imbalance, and summarizing results based on metrics we deem important. Where possible, we will perform the exact same exercise here, except we will use the "tidymodels" suite of packages to do so.
There are a few sources from which this tutorial draws influence and structure.
- "Tutorial on tidymodels for Machine Learning": hansjoerg.me/2020/02/09/tidym...
- "Tidymodels: tidy machine learning in R": www.rebeccabarter.com/blog/202...
- "Caret vs. tidymodels - comparing the old and new" by Konrad Semsch: konradsemsch.netlify.app/2019...
- "Tidy Modeling with R" by Max Kuhn and Julia Silge: www.tmwr.org/
- Recursive feature elimination example by Max Kuhn: github.com/stevenpawley/recip...
- Documentation for "stacks": stacks.tidymodels.org/article... - Věda a technologie
Awesome 👏, thanks 🙏
Many thanks, Richard, can I request if you can explain how we go about a continuous variable e.g., weight, age, etc instead of the discrete variable as the outcome. greatly appreciated.
Your caret tutorials were a pleasure, but this is a terrible slog. The script is nonsensical. Your tutorials are always excellent but I can see even you are jumping through more hoops in an attempt to make it more understandable. I'm going to see it through to part 3 though just to get the full picture.
I'm sorry to hear that! I really like tidymodels and for me it took a little while, but once it did it really clicked - but no package is for everyone.
@@RichardOnData indeed, I'll keep an eye on how tidymodels develops. I inadvertently got a lot more from these caret/tidymodels videos than just the package. Some background, I've been using R for a few years. When I started I was overwhelmed by the package choice, so when faced with the "tyranny of choice" I chose nothing, I did everything in Base R wherever possible (and stubbornly so). I only found your videos because I wanted to know more specifically about caret (as other ML practitioners I know were using it). However, your enthusiasm for the tidyverse, made me give your tidyverse series of videos a go (btw excellent!, in your next life/career you should be a teacher). In short, I came here for the caret, but left with the tidyverse. Many thanks