6.6 Meeting Videos
6.6.1 Cohort 1
Meeting chat log
00:15:40 Tan Ho: YESSS
00:15:58 Tan Ho: (@ the drake meme)
00:21:51 Tyler Grant Smith: space advantage
00:23:32 Asmae : what's that
00:24:28 Jim Gruman: http://www.feat.engineering/categorical-trees.html on dummies or no-dummies with trees
00:24:37 Tony ElHabr: amazing meme
00:24:48 Tony ElHabr: I am physically applauding
00:25:03 Scott Nestler: Some additional discussion regarding fit_xy() at tidyverse.org/blog/2019/04/parsnip-internals, related to possible range of mtry variables when you don't know number of predictors before recipe is prepped.
00:25:56 Scott Nestler: Is there a typo in the book after tables toward end of 7.1 where it talks about common argument names? It mentions num_n but I think they meant min_n. What do others think? Or is num_n actually used?
00:26:30 Conor Tompkins: Scott I think Jon made a PR to fix that typo
00:26:53 Scott Nestler: Thx. I hadn't checked yet. Just caught it in a quick read right before we started.
00:31:15 Scott Nestler: There are actually 30 different model types and engines at https://www.tidymodels.org/find/parsnip/ that work with parsnip.
00:43:55 Tony ElHabr: yay volunteers
00:44:48 Andy Farina: Thanks Jordan, excellent presentation6.6.2 Cohort 2
Meeting chat log
00:12:53 shamsuddeen: https://torch.mlverse.org
00:13:21 shamsuddeen: Book: https://mlverse.github.io/torchbook_materials/
00:16:18 Janita Botha: I'm in the meeting twice because my sound is not working this morning
00:26:10 August: fit_xy doesn't do the contrast coding automatically, fit does.
00:28:04 Luke Shaw: Is fit_xy a 'safer' option?
00:28:33 rahul bahadur: @August. If you were to have categorical predictors in `lm()` would fit_xy() be able to model it? `lm()` automatically does contrast coding internally
00:30:55 Luke Shaw: https://xkcd.com/927/
00:33:10 Amélie Gourdon-Kanhukamwe (she/they): Nice one Luke!
00:34:44 rahul bahadur: I hope tidymodels doesn't end up becoming just another standard :P
00:34:50 August: I'm not sure about safer, it depends I suppose if you have done a recipe with the coding before hand.
00:35:39 Kevin Kent: I wonder if future packages that tidy models references will be built with a tidy models integration in mind. Instead of tidy models having to do so much translation
00:36:34 Luke Shaw: Thanks August. Think I need to get my hands dirty using it to really help me understand!
00:37:34 Luke Shaw: Yeah I'm only joking with the xkcd but does feel relevant. Kinda hoping tidymodels can be the one standard I get v. familiar with and has a long life
00:40:30 Kevin Kent: I think modeltime is an example of a modeling package done with tidymodels in mind. Kind of designed first for tidy models instead of having to do a bunch of translation to fit into the framework. Hopefully more do development like this
00:41:21 Kevin Kent: But I guess that one is doing a similar job to tidymodels in the sense of pulling in functionality from other packages as engines
00:41:45 August: Absolutely.
00:42:44 August: btw I have mostly used fit() rather than fit_xy() although the latter by be preferable
00:42:54 Kevin Kent: I love these templating packages..excited to use them
00:43:18 Kevin Kent: rstudio add in looks fantastic
00:44:23 rahul bahadur: Yes, fit_xy(), though requires more code, would not lead to unknown behaviour. I believe.
00:44:41 Kevin Kent: Yeah less “magic” happening
00:46:19 Luke Shaw: This feels very scary about using "fit": "If the data were preprocessed in any way, incorrect predictions will be generated (sometimes, without errors)."
00:47:56 Luke Shaw: ^^^ that's not about the function fit() - my bad
00:48:09 Kevin Kent: Or if you accidentally read in a numeric column as categorical and then run fit it won’t throw an error even though its not meaningful as a categorical variable6.6.3 Cohort 3
Meeting chat log
00:15:22 Ildiko Czeller: > linear_reg()
Linear Regression Model Specification (regression)
> linear_reg() %>% set_mode("classification")
Linear Regression Model Specification (classification)
00:17:00 Edgar Zamora: https://www.tidymodels.org/find/parsnip/
00:22:51 Toryn Schafer (she/her): model.matrix
00:26:52 Ildiko Czeller: Hi Daniel! good to see you
00:27:14 Daniel Chen: hello hello!
00:29:02 Daniel Chen: I forgot there was an article/help page of all the supported engines
00:30:04 Daniel Chen: ooh they moved the things around: this is the page to help you find arguments: https://www.tidymodels.org/find/parsnip/
00:30:40 Daniel Chen: that page has the model/engine parsnip param name and original param name
00:31:03 Ildiko Czeller: thanks!
00:34:36 Daniel Chen: you just need to be careful with the column names after `tidy` since they don't always mean the same thing for each model output.
but it makes the column names consistent so you can combine tables
00:45:18 jiwan: Error in parsnip::rand_forest(verbose = TRUE) :
unused argument (verbose = TRUE)
00:45:48 Daniel Chen: unused arguments usually get ignored. maybe you get a warning?
00:54:29 Daniel Chen: bye all!6.6.4 Cohort 4
Meeting chat log
00:56:18 Laura Rose: https://emilhvitfeldt.github.io/ISLR-tidymodels-labs/linear-model-selection-and-regularization.html#forward-and-backward-stepwise-selection
00:58:07 Ryan Metcalf: https://emilhvitfeldt.github.io/ISLR-tidymodels-labs/index.html
00:58:52 Ryan Metcalf: https://avehtari.github.io/ROS-Examples/
01:01:31 Federica Gazzelloni: https://arxiv.org/pdf/1502.06988.pdf