We can use data preprocessing recipes to implement dimensionality reduction and understand how features of songs are related to their performance on the Billboard Top 100 chart. Check out the code on my blog: juliasilge.com...
This is what I've been looking for thank you very much. Now to figure out how to quantify which approach provides the most accurate model performance. Thanks again.
That's a great video, thank you for uploading this video :D Hope someday, you can make a video about time series using tidymodels with hyperparameter tuning
Excellent tutorial and demonstration! However, there is one thing I don't really get. Why would one define an outcome variable for PCA? I mean isn't the point of this type of models to reveal hidden dimensions that express the variability of the data without having an outcome variable?
Oh for sure, it is not necessary to specify an outcome. You can see how to set up a recipe step with no outcome in the examples here: recipes.tidymodels.org/reference/step_pca.html#ref-examples The reason I included an outcome here was to show how you could use dimensionality reduction as a preprocessing step before fitting models, like this: www.tmwr.org/dimensionality.html#bean-models
@@JuliaSilge Thank you for the links! Chapter 16 of the book uses the "class" variable as the outcome variable for all four types of PCA that are exemplified and it was not clear to me whether I would need to add an outcome variable while preparing the recipe. You reply here clears that out. In the mean time, I also watched yesterday an earlier video of yours that uses "~." for ignoring the outcome variable, while conducting PCA. Keep up the good work!
I wouldn't say "need to" per se; you can read more here: stats.stackexchange.com/questions/92791/why-does-sphericity-diagnosed-by-bartletts-test-mean-a-pca-is-inappropriate
@@JuliaSilge Thanks Julia! I'll take a look. Do you test these assumptions when diving into PCA? Or what approach do you take if that makes sense? So many nuances to these methods/assumptions can definitely bog down some steps.
I am only able to set the resolution to 360p. Do you still have the original recording at a higher resolution? It is difficult to read the text in the IDE. thak you for the excellent tutorials. I know it is a lot of work.
@@JuliaSilge By definition PCA is a method for quantitative variables. My question is about handle tabular data to apply multidimensional reduction as PCA, PLS or UMAP.
@@guberney I'm not sure what you mean here; "tabular data" typically means data arranged in a table form with rows and columns like what I have used here, like what you would find in a CSV or database or an R dataframe. Maybe you mean something else?
@@guberney Ah gotcha. You can use `step_dummy()` to create dummy/indicator variables for any nominal/factor/string variables, before normalizing: recipes.tidymodels.org/reference/step_dummy.html