# 2021-05-25 Preview with Asmae Toumi
00:07:45 tan-iphone: timed with an audience
00:08:00 tan-iphone: did you get the cursive IDE set up?
00:11:00 Tony ElHabr: yup! victor mono is the secret
00:18:44 tan-iphone: foreach
00:18:47 tan-iphone: or future or something
00:18:53 Marschall Furman (he/him): parallel package
00:19:55 Jim Gruman: all_cores <- parallelly::availableCores(omit = 1)
future::plan("multisession", workers = all_cores) # on Windows
00:47:55 tan-iphone: "you'll learn when you grow up"
# 2021-06-01 Preview with Tony ElHabr
00:08:56 Jon Harmon: Like tensorflow but very few insane installation issues: https://github.com/mlverse/torch
00:23:46 Daryn Ramsden: Is it multinom_reg because there’s a logistic regression under the hood?
00:25:18 Jon Harmon: I think so, as far as google is telling me.
00:27:19 Asmae Toumi: Max’s quote: “One other note that is in the upcoming ensemble chapter about racing and stacking... the stack can only use the candidates that had the complete set of resamples done. With racing, it works fast by avoiding using all of the resamples for candidates that show poor performance. If the racing method quickly gets down to one tuning parameter set, that is the only candidate that can be added to the data stack (since it has the whole set of resamples.”
00:27:26 Pavitra: Why is there a rank jump from 2 to 12?
00:27:51 Jon Harmon: By model. It's showing the top 2 of each.
00:28:00 Daryn Ramsden: ohhhh
00:28:35 Jon Harmon: Er not by model per se but by... some combo....
00:28:41 Daryn Ramsden: So models ranked 3-11 are also rf?
00:31:58 Asmae Toumi: i’d be curious to see the stacks results without tune race anova
00:33:16 Jon Harmon: beepr always reminds me of this silliness that I had running a year and a half ago or so when Max had mentioned {tune} somewhere but the repo was still private... and it exactly solved something I was working on so I wanted to know as soon as it was public:
tune_check <- httr::GET(
url = "https://github.com/tidymodels/tune"
)
while (httr::status_code(tune_check) == 404) {
Sys.sleep(15*60)
tune_check <- httr::GET(
url = "https://github.com/tidymodels/tune"
)
}
beepr::beep(8)
00:58:21 Asmae Toumi: Tony hive forever
00:58:48 Asmae Toumi: Can you send the code and data? I want to understand it better too
01:00:11 Pavitra: Ledell
01:00:13 Asmae Toumi: Erin LeDell is a legend
01:00:22 Asmae Toumi: It was so funny
01:01:53 Pavitra: Thank you Tony!! This was awesome
01:01:54 Jim Gruman: thank you!!
01:02:00 Andrew G. Farina: Thank you Tony!
# 2021-10-05 Presentation with Simon Couch
00:18:39 Jon Harmon (jonthegeek): It's fun watching Simon's reactions to Asmae's presentation
00:25:40 Tony ElHabr: “the literature says” = max says
00:26:00 tan: "the tidyverse says" = Hadley says
00:29:06 Jon Harmon (jonthegeek): Current {stacks} version has elasticnet as default, not just lasso, Max! <- note to the future
00:49:53 Simon Couch: https://github.com/tidymodels/stacks/issues/91
00:52:28 Bryan Shalloway: There is parsnip::null_model() that you can use for comparing against the mean... not sure if it's possible to specify something *slightly* more sophisticated there...
00:52:47 tan: itsmesleepknee
00:52:55 Jim Gruman: thank you Asmae and Simon!!!
00:54:47 Tony ElHabr: See y’all next time… whenever that is
00:56:55 Pavitra-Dallas: Thanks asmae and Simon 🤗🤗🤗
00:59:13 tan: thanks folks!
Meeting chat log: SLICED Discussion
00:37:14 Bryan Shalloway: Curious if test set is a random sample? Or if it is segment time?
20.7.2 Cohort 3
Meeting chat log
00:11:22 Daniel Chen: hi. sorry i'm late
00:12:14 Federica Gazzelloni: hello
00:14:27 Daniel Chen: it's retraining all the indibidual models again?
00:14:32 Daniel Chen: before adding the weights?
00:19:51 Federica Gazzelloni: hello Jiwan you are stack 🙂
00:45:17 Daniel Chen: lm_form_fit %>% extract_fit_engine() %>% vcov()
#> (Intercept) Longitude Latitude
#> (Intercept) 212.621 1.6113032 -1.4686377
#> Longitude 1.611 0.0168166 -0.0008695
#> Latitude -1.469 -0.0008695 0.0330019
00:45:23 Daniel Chen: https://www.tmwr.org/models.html#use-the-model-results