dlab-berkeley/r-machine-learning-legacy

D-Lab's 6 hour introduction to machine learning in R. Learn the fundamentals of machine learning, regression, and classification, using tidymodels in R.

D-Lab R Machine Learning with tidymodels

This repository contains the materials for D-Lab's R Machine Learning with tidymodels. Prior experience with the concepts in R Fundamentals and Data Wrangling and Manipulation in R is assumed.

Workshop Goals

In this workshop, we provide an introduction to machine learning algorithms by making use of the tidymodels package. First, we discuss what machine learning is, what problems it works well for, and what problems it might work less well for. Then, we'll explore the tidymodels framework to learn how to fit machine learning models in R. Finally, we will apply the tidymodels framework to explore multiple machine learning algorithms in R.

By the end of the workshop, learners should feel prepared to explore machine learning approaches for their data problems.

Familiarity with R programming and data wrangling is assumed. If you are not familiar with the materials in Data Wrangling and Manipulation in R, we recommend attending that workshop first. In addition, this workshop focuses on how to implement machine learning approaches. Learners will likely benefit from previous exposure to statistics.

Installation Instructions

We will use RStudio to go through the workshop materials, which requires the installation of both the R language and the RStudio software. Complete the following steps:

1. Download R: Follow the links according to the operating system that you are running. Download the package, and install R onto your computer. You should install the most recent version (at least version 4.0).

2. Download RStudio: Install RStudio Desktop. This should be free. Do this after you have already installed R. The D-Lab strongly recommends an RStudio edition of 2022.02.0+443 "Prairie Trillium" or higher.

3. Download these workshop materials:

• Click the green "Code" button in the top right of the repository information.
• Click "Download Zip".
• Extract this file to a folder on your computer where you can easily access it (we recommend Desktop).

4. Optional: if you're familiar with git, you can instead clone this repository by opening a terminal and entering git clone [email protected]:dlab-berkeley/R-Machine-Learning.git.

5. Be sure to run the install.R script in the repository so that all necessary packages are installed.

Is R Not Working on Your Laptop?

This workshop makes use of many packages within the R ecosystem. For that reason, we recommend using R on your local machine.

If you do not have R installed and the materials loaded on your workshop by the time it starts, we strongly recommend using the UC Berkeley DataHub to run the materials for these lessons. You can access the DataHub by clicking the following button:

Some users may have to click the link twice if the materials do not load initially.

The DataHub downloads this repository, along with any necessary packages, and allows you to run the materials in an RStudio instance on UC Berkeley's servers. No installation is needed from your end - you only need an internet browser and a CalNet ID to log in. By using the DataHub, you can save your work and come back to it at any time. When you want to return to your saved work, go straight to DataHub, sign in, and click on the R-Machine-Learning folder.

If you don't have a Berkeley CalNet ID, you can still run these lessons in the cloud, by clicking this button:

If you are loading Binder with this repository for the first time, it may take a few minutes to set up. Binder operates similarly to the D-Lab DataHub, but on a different set of servers. By using Binder, however, you cannot save your work.

Run the Code

Now that you have all the required software and materials, you need to run the code:

1. Launch the RStudio software.

2. Use the file navigator to find the R-Machine-Learning folder you downloaded from Github.

3. Open up the file corresponding to the part of the workshop you're attending.

4. If necessary, run install.R to make sure the requisite packages are installed. This should not be necessary on Binder.

5. Place your cursor on a given line and press "Command + Enter" (Mac) or "Control + Enter" (PC) to run an individual line of code.

6. The solutions folder contains the solutions to the challenge problems.

Additional Resources

This workshop draws heavily on the following resources:

• Tidy Modeling with R by Max Kuhn and Julia Silge.

• An Introduction to Statistical Learning by James, Witten, Hastie, and Tibshirani

• "Machine Learning Methods Economists Should Know About" by Athey and Imbens

• Feature Engineering and Selection: A Practical Approach for Predictive Models by Kuhn and Johnson (2019)

• The Elements of Statistical Learning by Hastie, Friedman, and Tibshirani

• Deep Learning by Goodfellow, Bengio, and Courville

Other D-Lab R Workshops

Basic Competency

• R Data Wrangling
• R Graphics with ggplot2
• R Functional Programming
• Project Management in R
• Geospatial Fundamentals in R with sf
• Census Data in R

Intermediate/Advanced Competency

• Unsupervised Learning in R
• Introduction to Deep Learning in R
• Fairness and Bias in Machine Learning
• R Package Development

Contributors

• Alex Stephenson
• Pratik Sachdeva
• Hero Ashman

Previous iterations of D-Lab's Machine Learning with R were created by:

• Chris Kennedy
• Evan Muzzall
• Jae Yeon Kim
• Christopher Hench

r-machine-learning-legacy's Issues

Error producing metrics and confusion plot

In lines 251-256 of 04-decision-trees.rmd these two plots run into errors. I think it's because tree_fit_viz_metr and tree_fit_viz_mat are not defined already. Here is the relevant code:

`# Metrics
(tree_fit_viz_metr + labs(title = "Non-tuned")) / (visualize_class_eval(tree_fit_tuned) + labs(title = "Tuned"))

Confusion matrix

(tree_fit_viz_mat + labs(title = "Non-tuned")) / (visualize_class_conf(tree_fit_tuned) + labs(title = "Tuned"))`

update project file --> docs folder doesn't exist

'docs' folder doesn't exist in the R-Machine-Learning-main project, so there's an error that pops up when you open it. doesn't impact running code

Stacks Still broken

Error message

Error: The inputted candidates argument was not generated with the appropriate control

• settings. Please see ?control_stack

Add more intuitive examples for tree, bagging, boosting, and ensemble models

Typo in Part3

Line 70 in Part3.md should be set_mode rather than set_model.

Clarify installation workflow

From the instructions it is unclear which step is suppose to be performed first:

1. The first line in Part1 renv::init()
2. The 5th step in installation instructions install.packages(c("tidyverse", "tidymodels", "here","pROC","glmnet", "ranger", "rpart", "xgboost","rpart.plot", "doParallel", "palmerpenguins", "ISLR2", "klaR", "stacks"))

adding a summary of packages and functions

It would be helpful to add a short summary of the packages and functions introduced with their respective uses at the end of the third module (pre-processing)!

add codes for visualization in 05_more_models

To show if balances are different for default status:
plot(default$balance ~ default$default)

Include plot visualizing bias-variance trade off in 04_regularization.Rmd

Under the section Regularization under linear regression --> Ridge Regression

Solutions for 04_regularization

Solutions for 04_regularization should include:

Import data

penguins <- palmerpenguins::penguins %>%
filter(!is.na(bill_length_mm))

Set seed

set.seed(23)

Perform split

penguin_split <- penguins %>% initial_split(prop = 0.80)
penguins_train <- training(penguin_split)
penguins_test <- testing(penguin_split)

Participants will run into error if null values in bill_length_mm are not dropped

Add stacks to the 07-ensembles note

Binder link option doesn't give Rstudio environment to run code

In the installation instructions the step:

If you don't have a Berkeley CalNet ID, you can still run these lessons in the cloud by clicking this button: Binder_link_here

This leads you to a Jupyter environment that cannot run code in the .Rmd lesson files.

I suggest we remove this option.

Typo error in Part 3 .rmd random forest section

on line 345, there is a typo where instead of "recipe" the function says "ecipe".

Error in Part 3 .rmd, Ensembles Section, cannot find fit_members() function

On lines 381-389, R cannot find fit_members() function unless explicitly installing stacks package. We might need a section for this in the notebook.

Workshop Title

Workshop title should be "R-Introduction-to-Machine-Learning-with-tidymodels"

Random Forest cross validation currently failing

This needs to be fixed on its own and also because the ensembles model depends on it.

Confusion over new ridge penalty plot

The labels were backwards so it looked like as the penalty got larger, the coefficients get larger. The reverse is true.

recommendations for additional restructuring

I delivered the workshop with these current materials, and things went pretty smoothly, but it's too much material for two 3 hour workshops.

I'd recommend splitting this workshop into either 3 or 4 2-hour workshops. This could look something like:

• Part 1: introduction and regression
• Part 2: preprocessing
• Part 3: regularization and cross-validation
• Part 4: more models

Perhaps Parts 1/2 could be merged into a single workshop.

I'd also recommend beefing up the "more models" section to more appropriately fill a 2 hour slot: more details on logistic regression, naive bayes, and random forest

Error in Part 3 .rmd, random forest section, R saying not all variables are present in supplied training set

On lines 357-364, R runs into an error where it says that not all variables are available despite it seeming like they are.

"Error: Not all variables in the recipe are present in the supplied training set: 'student', 'balance', 'income', 'default'.
Run rlang::last_error() to see where the error occurred."

Part 3 .rmd, issue with naive_Bayes function

Tidyverse cannot find the naive_Bayes function, leading to an error in the Part 3 .rmd file on lines 300-305.

Typo: Fitting best model on training dataset

In lessons/Part3.Rmd, it's mentioned that the finalized LASSO model is fit with the training set, but in Line 158, it seems like a typo that it's fitted with data_test instead of data_train.

https://github.com/dlab-berkeley/R-Machine-Learning/blob/a12e89b65cbfd1356d2df04c736d4b2a914820fe/lessons/Part3.Rmd#L158

additional package to install for 04_regularization

install.packages("glmnet")

04.Decision_Trees.RMD `makeCluster()` function error

Running the part 4 .RMD locally, I'm getting an error on line 197 with:

cl <- makeCluster(detectCores() - 1)
registerDoParallel(cl)

I believe this function optimizes a local environment for heavier computation? I've dug into the docs and arguments get deep into CS, beyond my scope at this time. It's possible to continue downstream in the workflow without these lines being run, which is what I've done so far.

Issue with Part 3 .rmd, typo in Linear Regression Workflow section

on line 67, set_engine has an extraneous "l" which leads to an error

Standardize spacing around piping operators

Pipes (%>%) in the lessons do not consistently have spaces before them. The tidyverse style guide calls for spaces before pipe operators, which we should adhere to.

Minor word edits and correction to challenge in 04_regularization.Rmd

Use `renv` to ensure that attendees have the required packages

While the README contains instructions on installing the necessary packages for the workshop, it might be useful to have an renv setup so that users could simply run a renv::init() at the beginning of the first notebook in order to make sure all necessary packages are installed.

Add datahub and binder links and icons

@pssachdeva I just noticed this repo is missing binerhub links in the README: https://github.com/dlab-berkeley/R-Machine-Learning/blob/ee158d8506d68e72498a36d7484a403ed5b4b506/README.md?plain=1#L81-L87

Any chance you could add it and test it in time for today's workshop at 10am since we have folks from partner organizations who may need a binderized version?

Also, it would be great if you can add the iconified buttons like what's shown in R Fundamentals:

Thanks!

Hard stop errors when running .RMD(s) on my local environment.

• Overview-1.RMD
  needed to:

install.packages(‘rlang’)

before installation of all packages chunk, or else ggplot2 would not load.

• 04- Decision Trees, ln. 196
  Error in makePSOCKcluster(names = spec, ...) : Cluster setup failed. 4 of 4 workers failed to connect.

Error did not break the .RMD workflow.

• 05 - Random Forest
  update_model() function, same error as above

• 06- xgboost
  update_model(), same error, ln. 351

• 09 - hlclust

ln. 37:
Error: Problem with mutate() input ..1. x there is no package called ‘BBmisc’ℹ Input ..1 is across(is.numeric, BBmisc::normalize)

`install.packages("BBmisc")

library(BBmisc)

and np.

 

_

Typo: 03-Lasso, ln: 254 "initial

Typo in Part 2 section on `parsnip`

In Part2.md, line 122, the linear_reg model is missing parentheses when passing into the stan engine.

Simplify and update renv package installation process

The renv can be simplified to remove unnecessary packages (to speed up (re-)installation) and replaced with an exact step-by-step process for people to follow. Some participants struggled with package installation as it is currently written.

https://github.com/dlab-berkeley/R-Machine-Learning/blob/ee158d8506d68e72498a36d7484a403ed5b4b506/README.md?plain=1#L43-L75

Also, install_github() should be replaced with standard install.packages() where possible, e.g. remotes::install_github("tidymodels/discrim") can just be install.packages("discrim") since the standard v1.0.0 package has been released now.

This caused additional issues because it assumes they have already run install.packages("remotes") because it is not explicitly stated in the instructions, so add that if there's a good reason to keep any install_github calls around.

Participant Instructions to run 01_overview.RMD

It may help students and instructors if we include the "Participants Instructions" closer to the top of the read me, in large script with a message something like:

"If you are working in a local computing environment, be sure to prepare for the workshop by running all the code in the 01.Overview.RMD before our first meeting. Doing this could take a few minutes, and is needed for you to run the workshop's code on your own computer."

Added slides for tidymodels

Part 3 .rmd, issue with setting seed.

On line 216, R says that the seed is not a valid integer. Reducing the seed to four numbers fixes this issue.

01-overview.Rmd doesn't install missing packages

01-overview.rmd doesn't seem to install missing packages by default. I had to manually install tidymodels and and ck37r. Not sure if this is an issue on my end or with the code block.

Datahub-Rstudio package loading issues

When I open this workshop in datahub via the gitpuller, and run the first line:

renv::init()

I get the error message:

> knitr::opts_chunk$set(echo = TRUE)
Error in loadNamespace(x) : there is no package called ‘knitr’

When I try to install knitr manually, it takes a very long time.

Little stuff!

Typo: 03-Lasso, ln: 254
"initial"

Possible Typo: set_model => set_mode

In lessons/Part3.Rmd, I don't think parsnip::set_model exists. Perhaps parsnip::set_mode should be used.

https://github.com/dlab-berkeley/R-Machine-Learning/blob/d120942132d6f7bdbd9c6d6c59cdad72db46bcb9/lessons/Part3.Rmd#L78

Part 2 .rmd, error with R chunk code

Lines #315-340, this line of code should be in an R chunk but is not.

Undefined function: "unregister_dopar"

unregister_dopar doesn't seem to be a function from any loaded library.

Error in unregister_dopar() : could not find function "unregister_dopar"

https://github.com/dlab-berkeley/R-Machine-Learning/blob/5f627e0268ed7ac2998f96bea28ab1e40e10886f/lessons/Part3.Rmd#L373

Perhaps that function came from this Stack Overflow answer.

Repo currently using Iris dataset