Chapter 12 Appendix

12.1 A. Package Reference

12.1.1 Core Tidyverse

Package Purpose Used In
dplyr Data manipulation All chapters
tidyr Data reshaping Chapters 3, 5, 7
readr Fast I/O All chapters
purrr Functional programming Chapters 6, 8
stringr String manipulation Chapters 3, 5
lubridate Date/time handling Chapters 4, 6, 7
ggplot2 Visualization Chapters 3, 5, 8

12.1.2 Modeling

Package Purpose Used In
tidymodels ML framework Chapters 3, 4
xgboost Gradient boosting Chapters 3, 4, 6
prophet Time series Chapter 4
modeltime Time series workflows Chapter 4
cluster Clustering algorithms Chapter 5
factoextra Cluster visualization Chapter 5
DALEX Model explainability Chapter 3
vip Variable importance Chapter 3
solitude Isolation forests Chapter 6
bayesAB Bayesian testing Chapter 8
pwr Power analysis Chapter 8
gsDesign Sequential testing Chapter 8

12.1.3 Applications

Package Purpose Used In
shiny Web applications Chapters 2, 4
shinydashboard Dashboard UI Chapter 4
plotly Interactive charts Chapters 4, 8
DT Interactive tables Chapter 4
plumber REST APIs Chapters 3, 6
blastula Email automation Chapter 7

12.1.4 Infrastructure

Package Purpose Used In
targets Pipeline orchestration Chapter 7
tarchetypes Pipeline helpers Chapter 7
DBI Database connectivity All chapters
RPostgres PostgreSQL driver All chapters
dbplyr Lazy SQL evaluation Chapters 3, 4
pool Connection pooling Chapters 2, 3
bigrquery BigQuery interface Chapters 3, 5
redis In-memory store Chapter 6
logger Structured logging All chapters
renv Reproducible environments Chapter 2
pins Artifact versioning Chapter 11

12.2 B. Function Quick Reference

12.2.1 Database Operations

tbl(con, "table_name")           # Create lazy reference
collect()                        # Execute and bring to R
show_query()                     # View generated SQL
inner_join(), left_join()        # Join operations

12.2.2 Tidymodels Workflow

recipe() %>% step_*()            # Feature engineering
workflow() %>% add_recipe() %>% add_model()  # Bundle recipe + model
tune_grid()                      # Hyperparameter search
select_best()                    # Choose optimal parameters
finalize_workflow() %>% fit()    # Train final model

12.2.3 XGBoost

xgb.DMatrix()                    # Create training matrix
xgb.train()                      # Train model
xgb.cv()                         # Cross-validation
xgb.importance()                 # Feature importance

12.3 C. SQL Translation Reference

Common dplyr to SQL translations:

dplyr SQL
filter(x > 5) WHERE x > 5
select(a, b) SELECT a, b
mutate(c = a + b) SELECT *, a + b AS c
group_by(x) %>% summarise(y = mean(z)) SELECT x, AVG(z) AS y FROM ... GROUP BY x
arrange(desc(x)) ORDER BY x DESC
left_join(b, by = "id") LEFT JOIN b ON a.id = b.id
head(10) LIMIT 10

12.4 D. Further Reading

12.4.1 Books

  • R for Data Science (Wickham & Grolemund) — Foundation
  • Tidy Modeling with R (Kuhn & Silge) — tidymodels deep dive
  • Feature Engineering and Selection (Kuhn & Johnson) — Feature craft
  • Designing Machine Learning Systems (Huyen) — MLOps architecture

12.4.2 Papers

  • “A Unified Approach to Interpreting Model Predictions” (Lundberg & Lee, 2017) — SHAP
  • “Planetary-Scale Prediction with XGBoost” (Chen & Guestrin, 2016)
  • “Forecasting at Scale” (Taylor & Letham, 2017) — Prophet

12.5 E. Glossary

A/B Testing: Controlled experiment comparing two variants to determine which performs better.

Class Imbalance: When one class in a classification problem has significantly more observations than others.

Data Drift: Change in the statistical properties of input data over time, degrading model performance.

Feature Engineering: The process of transforming raw data into features that better represent the underlying problem.

Hyperparameter Tuning: The process of finding the optimal configuration for a machine learning algorithm.

Lazy Evaluation: Delaying computation until the result is actually needed, common in database operations.

MLOps: Practices for deploying and maintaining machine learning models in production.

Removal Effect: In Markov chain attribution, the decrease in conversions when a channel is removed.

SMOTE: Synthetic Minority Over-sampling Technique for addressing class imbalance.

Stratified Sampling: Sampling method that ensures subgroups are represented proportionally.

Time Series Cross-Validation: Cross-validation that respects temporal ordering to prevent data leakage.

Thank you for reading Production Data Science with R. Now go build something that lasts.