Datasets¶

The course datasets live in ALL CSV FILES - 2nd Edition/ and are distributed by the textbook authors at statlearning.com for use with the book.

How the notebooks resolve data

The labs load datasets straight from the ISLP package wherever possible; the four the package does not ship (Advertising, Heart, Income1, Income2) stream from the book’s official site, and the bundled CSVs act as an offline fallback. You should never have to download anything by hand.

What’s bundled¶

File	Rows	Cols	Contents	Mainly used in
`Advertising.csv`	200	5	Sales against TV / radio / newspaper budgets	Ch 2–3
`Auto.csv`	397	9	Fuel economy and specs of 1970s–80s cars	Ch 3, 5, 8
`Bikeshare.csv`	8 645	16	Hourly bike rentals in Washington DC (counts)	Ch 4 (Poisson)
`Boston.csv`	506	14	Housing values in Boston suburbs	Ch 3, 8
`BrainCancer.csv`	88	9	Survival times of brain-cancer patients	Ch 11
`Caravan.csv`	5 822	86	Caravan-insurance purchases (highly imbalanced)	Ch 4
`Carseats.csv`	400	11	Child car-seat sales at 400 stores	Ch 3, 8
`Ch12Ex13.csv`	999	40	Gene-expression data for the Ch 12 exercise	Ch 12
`College.csv`	777	19	Statistics for 777 US colleges	Ch 6
`Credit.csv`	400	11	Credit-card balances and customer attributes	Ch 3, 6
`Default.csv`	10 000	4	Credit-card default (the classification workhorse)	Ch 4
`Fund.csv`	50	2 001	Monthly excess returns of 2 000 fund managers	Ch 13
`Heart.csv`	303	15	Heart-disease diagnosis	Ch 8
`Hitters.csv`	322	20	Baseball salaries and career statistics	Ch 6
`Income1.csv`	30	3	Income vs. years of education (simulated)	Ch 2
`Income2.csv`	30	4	Income vs. education and seniority (simulated)	Ch 2
`OJ.csv`	1 070	18	Orange-juice brand purchases	Ch 8–9
`Portfolio.csv`	100	2	Two asset returns — the bootstrap example	Ch 5
`Publication.csv`	244	10	Time to publication of clinical trials	Ch 11
`Smarket.csv`	1 250	9	Daily S&P 500 returns, 2001–2005	Ch 4
`Wage.csv`	3 000	11	Wages of Mid-Atlantic male workers	Ch 1, 7
`Weekly.csv`	1 089	9	Weekly S&P 500 returns, 1990–2010	Ch 4

Auto.data is the original whitespace-delimited version of Auto.csv, kept for the exercise that demonstrates reading a non-CSV file.

Loading a dataset¶

From the ISLP package

from ISLP import load_data

Boston = load_data("Boston")
Boston.head()

From the bundled CSVs

import pandas as pd
from pathlib import Path

DATA = Path("ALL CSV FILES - 2nd Edition")     # relative to the repo root
Boston = pd.read_csv(DATA / "Boston.csv", index_col=0)

Several files carry an unnamed index column (Boston, College, Heart, Income1, Income2, Bikeshare, BrainCancer, Publication, Fund) — pass index_col=0 for those.

Advertising (not in ISLP)

import pandas as pd

url = "https://www.statlearning.com/s/Advertising.csv"
Advertising = pd.read_csv(url, index_col=0)

Missing values

Auto contains a handful of rows with ? in horsepower; ISLP’s load_data drops them, so a CSV fallback should do the same (pd.read_csv(..., na_values="?").dropna()) for results to match the slides.

Attribution¶

The datasets are © the textbook authors and are distributed by them at statlearning.com for use with the book. They are bundled here only so the course runs offline — see Citation & licence.

Where to go next¶

Lab notebooks — where each dataset is loaded and used.
Python environment — how ISLP resolves the data.
Lecture slides — the figures computed from these files.