A reading companion for Econ 368 · Bates College · Winter 2027
Hitanshu Pandit · Department of Economics
For reading I recommend Aurélien Géron, Hands-On Machine Learning with Scikit-Learn and PyTorch (O’Reilly, 2025). This is not a replacement for it. It is my teaching notes written up, in the order we teach the course, with economic examples and Python you can run.
Use it for three things: to read the week’s material before Monday, to run the code (every block works in the course environment or Google Colab), and to work the exercises at the end of each chapter. Géron is where you go for the fuller treatment and more engineering detail; this is where you go for how we did it in class.
| Chapter | Covers | |
|---|---|---|
| Part I — Tools for economic data | ||
| 1 | What data science is for | description, prediction and causation; what “big” data changes |
| 2 | Python for economists | types, lists and dictionaries, functions, NumPy, a regression by hand |
| 3 | Reproducible projects with Git | folders, scripts, paths, commits, branches, pull requests |
| 4 | Wrangling, bigger data, and pictures | tidy data, pandas verbs, merges, missing values, Parquet/DuckDB, charts that inform |
| 5 | Mapping opportunity | the Opportunity Atlas, rank-rank mobility, geopandas, choropleths |
| 6 | Getting data from the web | HTTP, JSON, APIs and keys, HTML and scraping, the ethics |
| Part II — Experiments and uncertainty | ||
| 7 | A/B tests | randomization, difference in means, power, CUPED, peeking, many metrics |
| 8 | Resampling: simulation and the bootstrap | Monte Carlo, coverage, the bootstrap, permutation tests |
| Part III — Machine learning | ||
| 9 | An end-to-end machine learning project | the landscape, train/test, pipelines, cross-validation, the test set is sacred |
| 10 | Training linear models | OLS as a predictor, gradient descent, polynomials and splines, learning curves |
| 11 | Regularization and model selection | ridge, LASSO, elastic net, choosing the penalty by cross-validation |
| 12 | Classification | logistic regression, other classifiers, confusion matrices, thresholds and costs |
| 13 | Support vector machines | the widest street, soft margins, kernels |
| 14 | Decision trees | splits, impurity, depth, why trees overfit |
| 15 | Ensembles: forests and boosting | bagging, random forests, gradient boosting, importance and SHAP |
| 16 | Dimensionality reduction | the curse of dimensionality, PCA, a development index |
| 17 | Clustering | k-means, choosing k, hierarchical clustering, Gaussian mixtures |
| 18 | Neural networks | from logit to a neuron, backpropagation, training, when nets lose to trees |
| Part IV — Language | ||
| 19 | Text as data | regex, bag of words, tf-idf, dictionaries, text regression |
| 20 | Embeddings and topic models | topic models, word and sentence embeddings, similarity |
| 21 | Transformers, LLMs, and how they are told to behave | attention, pretraining and fine-tuning, constitutions, LLMs as annotators |
Every chapter ends with Economics reading: published papers in which economists use the chapter’s method, each with a question to think about.
Read with a laptop open. Every code block runs, top to bottom, and the numbers in the text are the numbers the code prints. Change something and run it again; that is how this material is learned.
The datasets repeat on purpose. Bay Area rents appear in Chapters 9, 10, 14 and 15; hotel bookings in Chapters 12 and 13; county childcare costs in Chapters 4 and 11. You are meant to notice that a new method is answering a different question about data you already understand.
Every number in this book is computed by the code in it. If a number here disagrees with a slide, the book is right and the slide is wrong. Please tell me in class or open an issue.
© 2026 Hitanshu Pandit. Text released under CC BY-NC-SA 4.0; code under the MIT license.