class: center, middle, inverse, title-slide # Yet Another rstudio::conf Recap ## Ann Arbor R Users Group ### Clayton Yochum<br><span class="citation">@claytonyochum</span> ### February 8, 2018 --- ## Overall Vibes -- - fully worth it -- - good food & well organized -- - Kraig got to fan out on Hadley -- - hex stickers!  --- ## Asynchronous Operations in Shiny -- - Joe Cheng, CTO @ rstudio & creator of Shiny -- - what happens if you run a model in an app, and someone else tries to connect? -- - new [`promises`](https://github.com/rstudio/promises) package coming - another weird pipe: `%...>%` - uses `futures` underneath - requires support in shiny itself, currently on [`shiny` branch](https://github.com/rstudio/shiny/tree/async) -- - not specific to shiny --- ## Functional Testing in Shiny -- - Winston Chang talking about [`shinytest`](https://github.com/rstudio/shinytest) -- - functional testing using a headless browser -- - less fragile than `RSelenium` -- - easy to use: record yourself clicking on stuff -- - replay from generated code -- - returns json & jpg -- - json plays nice with version-control and CI -- - already has an [rstudio page](https://rstudio.github.io/shinytest/articles/shinytest.html) --- ## Scaling Shiny -- - Sean Lopp, software dev @ RStudio -- - demonstrated scaling a shiny app to to 10,001 concurrent users! -- - built on AWS - multiple RStudio Connect servers on EC2 - ALB Load Balancer - Postgres DB's -- - other tools used - Prometheus & Grafana for metrics - Fabric (python) for orchestrating the whole thing -- - new, unreleased `shinyloadtest` tool to fake users - should be able to support ~1k on a laptop --- ## Drill-Down Reporting in Shiny -- - Barbara Borges Ribeiro (`pool`!) -- - didn't go, but sounded cool -- - ask Kraig --- ## Modeling in the Tidyverse -- - Max Kuhn, Applied Predictive Modeling & `caret` package - employee of RStudio since last fall -- - `modelr` is dead! -- - packages that already exist - [`rsample`](https://github.com/topepo/rsample) for setting up bootstrap, CV, etc. - [`recipes`]() for preprocessing (scaling, centering, etc.) - [`tidyposterior`](https://github.com/topepo/tidyposterior) for post-hoc analysis of model stats - [`yardstick`](https://github.com/topepo/yardstick) for computing model metrics - [`parsnip`](https://github.com/topepo/parsnip) for unified model interface --- ## Modeling in the Tidyverse -- - Max is clearly embracing the pipeline concept (smells like `sklearn`) - going beyond `recipes` -- - unified model interface means you specify a _type of model_ and a _compute target_ - e.g. `random forest` and `Spark` - don't worry about `randomForest` vs. `ranger` vs. `sparklyr::ml_random_forest` vs. ... -- - After he nails down some interface/syntax stuff, everything else will come quickly -- - Slides available on [Github](https://github.com/topepo/rstudio-conf) ([rawgit](https://cdn.rawgit.com/topepo/rstudio-conf/a6d9176bdc62f38c7d6773a8bcc2e6f1d4399536/2018/Modeling_in_the_Tidyverse--Max_Kuhn/Modeling_in_the_Tidyverse.html#1)) --- ## Zeallot -- - Nathan Teetor talking about his [`zeallot`](https://github.com/nteetor/zeallot) package -- - offers python-like value unpacking on LHS of assignment ```r library(zeallot) c(a, b) %<-% c(1, 2) c(a, b) %<-% list(1, "foo") c(a, c(b, c)) %<-% list(1, list("foo", "bar")) c(res, err) %<-% purrr::safely(log)("whoo") c(mpg, ...) %<-% mtcars c(., ., disp, ...) %<-% mtcars c(mpg, ...rest) %<-% mtcars ``` -- - easy to implement your own "destructor" for custom objects (demonstrated with `quosures`) -- - slides available via [Google Docs](https://docs.google.com/presentation/d/1MISSEW5-JIulvjsmGETsdLWrgXoncFhb2i2E65KXwi0/edit?usp=sharing) --- class: center, middle # Fin!