Appendix C — Dataset Descriptions

All data can be found in the book’s repo. Depending on when you access it, there may be more or less data available. We’ll try to clean it up to make it more clear eventually, but it’s easiest to use the code in the demonstrations to download the data directly.

C.1 Movie Reviews

The movie reviews dataset was a fun way to use an LLM to create movie titles and reviews in a specific way, as well as other features. With features in hand, we then generated a rating outcome with specific feature-target relationships. It has 1000 rows and the following columns:

  • title: The title of the movie
  • review_year: The year the review was written
  • age: The age of the reviewer
  • children_in_home: The number of children in the reviewer’s home
  • education: The educational level of the reviewer (Post-Graduate, Completed College, Completed High School)
  • gender: The gender of the reviewer (Male or Female)
  • work_status: The work status of the reviewer (Employed, Retired, Unemployed, Student)
  • genre: The genre of the movie
  • release_year: The year the movie was released
  • length_minutes: The length of the movie in minutes
  • season: The season the movie was released (e.g., Fall, Winter)
  • total_reviews: The total number of reviews for the movie
  • rating: The rating of the movie
  • review_text: The text of the review
  • word_count: The number of words in the review
  • review_year_0: The review year starting from 0
  • release_year_0: The release year starting from 0
  • *_sc: Scaled (standardized) versions of age, length_minutes, total_reviews, and word_count
  • rating_good: A binary version of rating, where 1 is a good rating (>= 3) and 0 is a bad rating (<3)

Link:

Repo File:

  • data/movie_reviews.csv
Table C.1: Movie Reviews Dataset (String)
variable empty n_unique
title 0.0 100.0
education 0.0 3.0
gender 0.0 2.0
work_status 0.0 4.0
genre 0.0 8.0
season 0.0 4.0
review_text 0.0 442.0
Table C.2: Movie Reviews Dataset (Numeric)
variable mean sd min med max
review_year 2015.8 5.1 2000.0 2017.0 2022.0
age 46.9 18.3 18.0 47.0 80.0
children_in_home 0.4 0.7 0.0 0.0 3.0
release_year 2008.1 9.6 1983.0 2010.0 2020.0
length_minutes 121.0 11.5 98.0 120.0 147.0
total_reviews 4921.7 2837.9 374.0 4464.0 9926.0
rating 3.1 0.6 1.0 3.1 5.0
word_count 10.3 5.1 2.0 9.0 32.0
rating_good 0.6 0.5 0.0 1.0 1.0

C.2 World Happiness Report

The World Happiness Report is a survey of the state of global happiness that ranks countries by how ‘happy’ their citizens perceive themselves to be. You can also find additional details in their supplemental documentation. Our 2018 data is from what was originally reported at that time (figure 2.2 in the corresponding report), and it also contains a life ladder score from the most recent survey, which is similar and very highly correlated.

The datasets contain the following columns:

  • country: The country name
  • year: The year of the survey
  • life_ladder: The happiness score
  • log_gdp_per_capita: The log of GDP per capita
  • social_support: The social support score
  • healthy_life_expectancy_at_birth: The healthy life expectancy at birth
  • freedom_to_make_life_choices: The freedom to make life choices score
  • generosity: The generosity score
  • perceptions_of_corruption: The perceptions of corruption score
  • positive_affect: The positive affect score
  • negative_affect: The negative affect score
  • confidence_in_national_government: The confidence in national government score
  • happiness_score: The happiness score
  • dystopia_residual: The dystopia residual score (difference from a ‘least happy’ country)

In addition, there are standardized/scaled versions of the features, which are suffixed with _sc.

Links:

Repo Files:

  • data/world_happiness_all_years.csv
  • data/world_happiness_2018.csv
Table C.3: World Happiness Report Dataset (All Years)
variable n_missing mean sd min med max
year 0.0 2014.2 4.7 2005.0 2014.0 2022.0
happiness_score 0.0 5.5 1.1 1.3 5.4 8.0
log_gdp_per_capita 20.0 9.4 1.2 5.5 9.5 11.7
social_support 13.0 0.8 0.1 0.2 0.8 1.0
healthy_life_expectancy_at_birth 54.0 63.3 6.9 6.7 65.0 74.5
freedom_to_make_life_choices 33.0 0.7 0.1 0.3 0.8 1.0
generosity 73.0 0.0 0.2 −0.3 0.0 0.7
perceptions_of_corruption 116.0 0.7 0.2 0.0 0.8 1.0
positive_affect 24.0 0.7 0.1 0.2 0.7 0.9
negative_affect 16.0 0.3 0.1 0.1 0.3 0.7
Table C.4: World Happiness Report Dataset (2018)
variable mean sd min med max
life_ladder 5.6 1.1 2.7 5.5 7.9
log_gdp_per_capita 9.3 1.2 6.6 9.5 11.5
social_support 0.8 0.1 0.5 0.8 1.0
healthy_life_expectancy_at_birth 64.7 6.8 48.2 66.7 75.0
freedom_to_make_life_choices 0.8 0.1 0.4 0.8 1.0
generosity 0.0 0.2 −0.3 0.0 0.5
perceptions_of_corruption 0.7 0.2 0.2 0.8 1.0
positive_affect 0.7 0.1 0.4 0.7 0.9
negative_affect 0.3 0.1 0.2 0.3 0.5
confidence_in_national_government 0.5 0.2 0.1 0.5 1.0
happiness_score 5.4 1.1 3.3 5.4 7.6
dystopia_residual 2.0 0.5 0.3 1.9 2.9

C.3 Heart Disease UCI

This classic dataset comes from the UCI ML repository. We took a version from Kaggle, and features and target were renamed to be more intelligible. Here is a brief description from UCI:

This database contains 76 attributes, but all published experiments refer to using a subset of 14 of them. In particular, the Cleveland database is the only one that has been used by ML researchers to date. The “goal” field refers to the presence of heart disease in the patient. It is integer valued from 0 (no presence) to 4. Experiments with the Cleveland database have concentrated on simply attempting to distinguish presence (values 1,2,3,4) from absence (value 0).

  • age: Age in years
  • male: ‘yes’ or ‘no’
  • chest_pain_type: ‘typical’, ‘atypical’, ‘non-anginal’, ‘asymptomatic’
  • resting_bp: Resting blood pressure (mm-Hg)
  • cholesterol: Serum cholesterol (mg/dl)
  • fasting_blood_sugar: ‘> 120 mg/dl’ or ‘<= 120 mg/dl’
  • resting_ecg: ‘normal’, ‘left ventricular hypertrophy’, ‘ST-T wave abnormality’
  • max_heart_rate: Maximum heart rate achieved
  • exercise_induced_angina: ‘yes’ or ‘no’
  • st_depression: ST depression induced by exercise relative to rest
  • slope: ‘upsloping’, ‘flat’, ‘downsloping’
  • num_major_vessels: Number of major vessels (0-3) colored by fluoroscopy
  • thalassemia: ‘normal’, ‘fixed defect’, ‘reversible defect’
  • heart_disease: ‘yes’ or ‘no’

Links:

Repo Files:

  • data/heart_disease_processed.csv
  • data/heart_disease_processed_numeric_sc.csv
Table C.5: Heart Disease UCI Dataset (String)
variable empty n_unique
chest_pain_type 0.0 4.0
fasting_blood_sugar 0.0 2.0
resting_ecg 0.0 3.0
exercise_induced_angina 0.0 2.0
slope 0.0 3.0
thalassemia 0.0 3.0
heart_disease 0.0 2.0
Table C.6: Heart Disease UCI Dataset (Numeric)
variable mean sd min med max
age 54.5 9.0 29.0 56.0 77.0
male 0.7 0.5 0.0 1.0 1.0
resting_bp 131.7 17.8 94.0 130.0 200.0
cholesterol 247.4 52.0 126.0 243.0 564.0
max_heart_rate 149.6 22.9 71.0 153.0 202.0
st_depression 1.1 1.2 0.0 0.8 6.2
num_major_vessels 0.7 0.9 0.0 0.0 3.0

C.4 Fish

This is a very simple dataset with a count target variable. It’s also good if you want to try your hand at zero-inflated models. The background is that state wildlife biologists want to model how many fish are being caught by fishermen at a state park.

  • nofish: We’ve never seen this explained. Originally 0 and 1, 0 is equivalent to livebait equals ‘yes’, so it may be whether the primary motivation of the camping trip is for fishing or not.
  • livebait: Whether or not live bait was used
  • camper: Whether or not they brought a camper
  • persons: How many total persons on the trip
  • child: How many children present
  • count: Number of fish caught

Link:

Repo File:

  • data/fish.csv
Table C.7: Fish Dataset (String)
variable empty n_unique
nofish 0.0 2.0
livebait 0.0 2.0
camper 0.0 2.0
Table C.8: Fish Dataset (Numeric)
variable mean sd min med max
persons 2.5 1.1 1.0 2.0 4.0
child 0.7 0.9 0.0 0.0 3.0
count 3.3 11.6 0.0 0.0 149.0