Thanks Andrew for your sharing! As a researcher in public health and medical sciences, I would probably be interested in doing cox regression models using tidymodels, as well as specifying splines for some of the terms, as these are rather common models in my field. Would you please make a video on this topic? Thanks!
I have limited experience with survival models (I once tried using them for time-to-arrival models for a project). However, there is an adjacent package to parsnip called censored (formally survsnip) that is made for censored data and also parsnip has survival_reg() now. I think I could make a brief video on survival analysis but I would want to wait for the censored package to be more developed. censored package: github.com/tidymodels/censored/ parsnip survival model: parsnip.tidymodels.org/reference/survival_reg.html
My settings in RStudio are: Rstudio Theme: Classic Zoom: 150% Editor Font Size: 10 Editor theme: Dracula I also have Rainbow parentheses enabled I also use R Markdown for most of my videos.
An extra question is why would we want to centralize or normalize the data in recipe? Are they making the models easier to converge? or to make the models fit faster? or other purposes?
I generally will normalize the data since usually I will experiment with other pre-processing/models that require it. In the video, I don't really need it but it is a habit that I have when modeling.
K-nearest neighbor, PCA, models that regularize. I think in general normalizing data will not hurt the model's performance and it helps prevent creating redundant pre-processing objects.