Information for the Applied Machine Learning workshop

workshop

#1

We will be sending out an email with the official information regarding the Applied Machine Learning workshop. This post we be used to supplement that information and enable a discussion about prerequisites, package installs, and any other questions that you have beforehand.

For this workshop, a GitHub repo will be provided via email about a week before the course. It will contain the code, slides, data, and everything else required for the workshop.

We will have RStudio Server Pro instances for those would like them. To help us plan, it would be great if you make a post below (or send me an email at max@rstudio.com or dm @Max ) if you don’t want to use a server (i.e. you will only run on your machine).

Local Installation Instructions

I’ll be using the the most recent version of R but I believe that anything > 3.4.1 should be fine. The package installation instructions are:

from_cran <- 
	c("AmesHousing", "broom", "caret", "devtools", "doParallel", "e1071", "earth", 
		"glmnet", "ipred", "klaR", "pROC", "rpart", "rsample", "sessioninfo", 
		"tidyposterior", "tidyverse", "yardstick")

install.packages(from_cran, repos = "http://cran.rstudio.com")

library(devtools)

install_github("topepo/recipes")

That install.packages command may additionally install over 100 more packages.

If you can install from source, it would be good (but not required) to get the development version of caret too:

install_github("topepo/caret", subdir = "pkg/caret")

To verify the installation, try running:

library(caret)
library(tidyposterior)
library(dplyr)
library(recipes)
library(AmesHousing)
ames <- make_ames()

RStudio Server Pro Instructions

We will send these out over email.

So… if you have any questions or issues, post them here!

EDIT: added a new verification case


#2

#3

This is awesome! Thanks @Max

I’m planning to run R on my machine.


#4

This is awesome. I am looking forward to the workshop


#5

I see they have the conference schedule posted, do you know where information is posted about class day schedules? i.e. what time will class start each day?


#6

@hlendway - I believe the detailed schedule is available in the conference app. I think the classes run 9 to 5 Wed and Thursday.


#7

Yes. We usually do 9:00-12:30 + 1:30-5:00 with two 30 minute breaks.


#8

Emailed question:

Will it be possible to cover during the workshop some of the deep learning libraries now available in R (such as Tensorflow and Keras)

No. I decided not to cover these they are more technically complex than the other models. I wanted to include Cubist models (since they work so well) but determined that it would take an inordinate amount of time to describe the model. Going through keras models would probably consume 1/4 of a day. Also, JJ and a few others will be giving talks on them too so more coverage would be redundant.

On the bright side, there are a lot of examples of keras code along with and a few different examples of using it with the newer tidy modeling packages.


#9

I may not be able to attend the Wednesday session until the afternoon (~12:00). I’m looking to rearrange things so as to join on time, but wanted to check whether tardiness will cause an issue or tremendous inconvenience (and whether I might mitigate such things).


#10

@Max - Looking forward to the workshop! How large are the datasets you plan to use in the tutorial? I think that will help me decide whether a RStudio server is necessary.


#11

@lawwu

> dim(okc_train)
[1] 4000   91
> dim(car_data)
[1] 3294   15
> dim(make_ames())
[1] 2930   81

On my 2015 mbp, some model fits take 4.5m and one takes 39m. However, we’ll use parallel processing and, on my laptop, that last one takes about 9m. The RSP instances are about 5% slower than this.


#12

Plan A is to use my own machine. I followed the instructions without issue and ran the verification steps make_ames().

Just curious, is it possible to run the server option as a Plan B should the un-foreseen get in the way of Plan A?

Really looking forward to it, btw.


#13

#14

#15

I wanted to take a minute and thank you for coming to the workshop. By far, you were the best group of students that I’ve had in a workshop. I really appreciate the enthusiasm (and capability) that you brought for two days.

One other note… a few people asked to see how I made the slides. They were created with Yihui’s xaringan and the stylings are a slightly modified version of Patrick Schratz’s Metropolis theme.

I added another file to the GitHub repo that is an Markdown document that has a few example slides.

Thanks again!


#16

Thanks @Max! This class was awesome! A definitely learned a lot. I’m thankful for all the materials, because I am now going to take it all apart on my own time: it is easier to learn this way for me.