Appendix A — Glossary

You’re halfway through a chapter, you hit a word like “kernel” or “PATH,” and nobody ever told you what it means. This page is for that moment. The terms are in alphabetical order, each with a short, plain definition and a pointer to the chapter that teaches it properly, so you can look one up and get straight back to what you were doing. When a chapter uses one of these words for the first time, it usually links here (a mention of a package in Chapter 14, for example, jumps straight to its entry below).

A lot of these words name layers of the software on your computer: the operating system at the bottom, then the file system, the shell, Python, its packages, and finally your own code. Together they’re sometimes called the technology stack. When something breaks, figuring out which layer is misbehaving is half the battle, and knowing its name makes the other half, searching for help or asking someone, a lot easier. Other words here are the vocabulary of looking after data so that someone else can find, trust, and reuse it: provenance, metadata, a license, a DOI.

Algorithmic audit

A close look at an AI or algorithmic system to find out whether it’s fair, safe, or accurate for the people it affects, including the failures nobody planned for. Who runs it matters: a first-party audit is done by the company that built the system, a second-party audit by the organization using it, and a third-party audit by an independent group with the access and the reason to find problems. Most audits in practice are first-party, so “we audited it” doesn’t by itself mean much. See Chapter 38, and Chapter 8 for why who holds the power matters.

API (application programming interface)

A published set of rules for how one program can ask another for data or services. In data work it usually means a web API: URLs that a service such as GitHub or the World Bank publishes for programs to call, which send back data (usually JSON) instead of a web page. Many APIs want an API key with each request, a secret that identifies you and belongs in an environment variable, never in your code. See Chapter 24 and Chapter 34.

Application

A program you open to get something done, like Microsoft Word, Excel, Google Chrome, or Apple Mail. Applications are built separately for each operating system, which is why a Mac app won’t run on Windows, and they need your permission to read and save files in some places. See Chapter 9.

Breakpoint

A line where a running program pauses so you can look around inside it with a debugger: the values of your variables, the chain of calls that got you there, and what happens as you run it one line at a time. You set one by writing breakpoint() in Python code or by clicking beside a line number in your editor; a conditional breakpoint pauses only when an expression you give it is true, such as row_id == 1047. See Chapter 6 and Chapter 12.

Channel

A named place conda downloads packages from. Anaconda and Miniconda start with Anaconda’s own defaults channel, while Miniforge starts with conda-forge, a community channel with far more packages. If conda says it can’t find a package, or two packages won’t install together, the channel is one of the first things to check. See Chapter 14.

Checksum

A short fingerprint computed from every byte of a file, like the 64-character string that sha256sum survey.csv prints. Change even one byte and the fingerprint changes, so a checksum you record when a file arrives lets you prove later that the file you analyzed is the one you downloaded, or that a big copy between computers arrived intact. See Chapter 30 and Chapter 11.

Command line interface (CLI)

A way of using your computer by typing commands instead of clicking, such as ls to list the files in a folder or python analyze.py to run a script. It looks bare at first, but it’s often faster than clicking, it can be scripted, and it’s the only option on many servers. On a Mac you get to it through the Terminal app; on Windows, through Windows Terminal running PowerShell, Command Prompt, or (the one the book recommends) WSL or Git Bash. See Chapter 11.

CSV (Comma-Separated Values)

A plain-text file format for tables: each row is a line, and the columns are separated by a delimiter, usually a comma (sometimes a tab or a semicolon). Because a CSV stores everything as text, the program that reads it has to guess what’s a number, a date, or a missing value, and it sometimes guesses wrong. See Chapter 20 for how to read one reliably.

Data dictionary

A small table that explains a dataset one column at a time: each column’s name, its type, what it means in plain words, its units, the values it’s allowed to take, and how missing values are marked. Keep it as a file next to the data, and someone else (or you, six months from now) can use the data correctly, and a short script can check new data against it. A codebook is the survey-research version, which adds each question’s exact wording and what its answer codes mean (1 = “strongly disagree”). See Chapter 30.

De-identification

Removing or disguising the details in a dataset that could reveal who a person is. Direct identifiers such as names, email addresses, and student IDs are the obvious ones, but indirect ones can single someone out in combination (a zip code, a birth date, and a gender together), which is how supposedly anonymous data keeps getting re-identified. The usual practice is to replace each person with a code and keep the key that links codes to people apart from the data, with fewer people able to open it; as long as that key exists, privacy laws such as the EU’s GDPR still count the data as personal. See Chapter 30 and Chapter 34.

DOI (Digital Object Identifier)

A permanent ID for a paper, a dataset, or a release of some code, like 10.1371/journal.pcbi.1005510. Put https://doi.org/ in front of it and you land on the work wherever it’s hosted now, so a DOI keeps working after a website reorganizes, which is why citation styles ask for one. Data repositories such as Zenodo give each deposit a DOI, and Zenodo can do the same for each GitHub release of your code. See Chapter 31 and Chapter 26.

Driver

A small program that lets the operating system talk to a piece of hardware, such as your Wi-Fi card, printer, graphics card, or trackpad. Your operating system keeps most drivers up to date through its own updates, so install drivers only from the operating system or the hardware’s maker, never from a “driver updater” site. See Chapter 9.

Environment

Everything your code runs inside of, beyond the code itself. In Python talk it usually means one Python interpreter and the packages installed for it, and you can have many on the same computer, one per project. More broadly, a program’s environment also includes its environment variables (like PATH) and the folder it was started from. When code works on one computer and fails on another, a different environment is the usual suspect. See Chapter 14 and Chapter 15.

Environment variable

A named setting, like PATH or HOME, that a running program carries with it; a program you start from a terminal gets a copy of the terminal’s. Many tools read their settings from environment variables, which makes one the usual home for an API key or password: your code mentions only the variable’s name, and each person who runs it supplies their own value. See Chapter 11 and Chapter 34.

File system

The part of your computer that stores your files and folders, keeps track of where each one is, and decides who’s allowed to read or change it. Every file has an address in it, called a path, like Documents/thesis/data/survey.csv. See Chapter 10.

Git

The most widely used version control system: a program that records the history of a project folder as a series of snapshots, called commits, so you can see what changed, go back to an earlier state, and combine work from several people. Git runs on your own computer. GitHub is a separate website that hosts copies of Git repositories and adds pull requests, issues, and reviews on top. See Chapter 31 and Chapter 32.

Graphical user interface (GUI)

The windows, icons, menus, and buttons you click to use a computer, as opposed to typing commands. Finder on macOS and File Explorer on Windows are GUIs for your file system. You’ll sometimes hear GUI pronounced “gooey.” See Chapter 11 for what typing commands can do that clicking can’t.

JSON (JavaScript Object Notation)

A plain-text format for structured data, built from { } objects (named fields) and [ ] lists that can nest inside each other, like {"name": "Ada", "courses": ["INFO 1201", "STAT 2010"]}. It’s what most web APIs send back, what many tools use for settings, and even what a Jupyter notebook is underneath. See Chapter 4 and Chapter 20.

Jupyter kernel

The separate program that actually runs the code in your Jupyter notebook; for a Python notebook, it’s a Python interpreter. The kernel remembers every variable you’ve created until it restarts, even ones whose cells you’ve since deleted, which is why a notebook can “work” only because of something still in memory. It also decides which Python (and so which installed packages) your code sees, so if an import fails in a notebook but works in the terminal, check the kernel first. See Chapter 16 and Chapter 15.

Large language model (LLM)

The kind of AI model behind ChatGPT, Claude, Gemini, and GitHub Copilot. It’s a neural network trained on huge amounts of text, and it writes by predicting, over and over, which small piece of text (a token) should come next. That makes it good at producing text that sounds right, which isn’t the same as text that is right. See Chapter 35 and Chapter 36.

Library

Code someone else wrote that adds abilities to a programming language. Python on its own covers the basics; libraries like numpy, pandas, and matplotlib add fast math, data tables, and plotting. In Python, “library” and “package” are used almost interchangeably. When a traceback shows a path containing site-packages/, that’s a library you installed. See Chapter 14.

License

The terms that say what other people may do with a piece of work: copy it, change it, share it, sell it. Without one, copyright law’s default applies and nobody else has clear permission to reuse what you’ve published, which is why a public repository needs a LICENSE file (choosealicense.com helps you pick one, such as MIT for code, or CC BY or CC0 for data and writing). It cuts both ways: the license on data you download decides whether you’re allowed to put it in a public repository or share it at all. See Chapter 30.

Markdown

A simple way to format plain text with ordinary characters: # for a heading, **bold** for bold, - for a list item, and backticks for code. The raw file stays readable, and tools turn it into a nicely formatted page. It’s how README files, GitHub issues, Jupyter text cells, and Quarto books like this one are written. See Chapter 4.

Matilda effect

The pattern of women’s contributions to research being under-cited and credited to someone else, often a male colleague. The historian of science Margaret Rossiter named it in 1993, after the suffragist Matilda Joslyn Gage, as a counterpart to the Matthew effect, in which people who already have credit get more of it. The same pattern shows up for scholars of color and scholars outside the most powerful institutions and countries. See Chapter 26 and Wikipedia’s article.

Metadata

Data about data: a description of a file or dataset rather than its contents. A file’s name, size, and last-modified date are metadata, and so are a commit’s author, date, and message; for a dataset, its provenance note and data dictionary are the metadata that let someone else find, understand, and cite it. Metadata can also say more than you meant to, like the author’s name stored inside a PDF you submit for anonymous review. See Chapter 30, Chapter 31, and Chapter 26.

Network drive

A folder on another computer, usually a department file server, that your computer connects to over the network and shows as if it were local: a mapped drive on Windows (\\server\share), or a server you connect to from Finder on a Mac (smb://server/share). Every read and write crosses the network, so it’s slow for big files and it vanishes when you go offline. See Section 10.7.

Notebook

A document that mixes code you can run, the output it produces (tables, charts, errors), and written notes, all in one file. The Jupyter notebook (a .ipynb file) is the most common kind in Python. Great for exploring, but the order you ran the cells in isn’t always the order on the page, so restart and run everything from the top before you trust the results. See Chapter 16.

Online-only file

A placeholder a sync client leaves on your disk to save space: the file’s name and icon are there, but its contents stay in the cloud until something opens it. OneDrive calls this Files On-Demand, and iCloud Drive does it when Optimize Mac Storage is on. It’s handy for documents, but code that reads an online-only file has to wait for a download, and fails if you’re offline. See Section 10.7.

Open access (OA)

Research that anyone can read for free online, with no subscription or paywall. Gold OA means the journal itself publishes the article openly, sometimes paid for by a fee from the author or their university; green OA means the author has posted a free copy somewhere else, such as arXiv or their university’s repository. See Chapter 25 and the Directory of Open Access Journals.

Operating system

The software between your apps and your hardware: macOS, Windows, and Linux are all operating systems. It launches and stops programs, runs the file system, keeps user accounts and their permissions separate, and talks to your keyboard, screen, Wi-Fi, and storage through drivers. See Chapter 9.

Package

Code that someone else has already written and published for you to install and use, such as numpy, pandas, or requests. The word has a second, narrower meaning in Python: a folder of your own .py files that you import as one unit, usually marked by an __init__.py file inside it. See Chapter 14 and Chapter 17.

Package manager

A tool that installs, updates, and removes packages for you, and works out which versions of everything they depend on can live together. pip and conda are the two you’ll meet most in Python data work, and uv is a fast newer one. See Chapter 14.

Parquet

A file format for tables that’s built for analysis. It stores data column by column, compresses it, and remembers each column’s type, so dates stay dates. You can’t read it in a text editor, but it’s usually much smaller and faster to load than the same data as a CSV. See Chapter 20.

PATH

An environment variable holding the list of folders your shell searches, in order, when you type a command like python. command not found means the program isn’t in any of those folders, and running the “wrong” Python usually means a different one comes first on the list. (Don’t confuse it with a lowercase path, the address of a file, like data/survey.csv.) See Chapter 11.

Personal data

Any information about a person who can be identified from it, on its own or combined with other information: a name, an email address, a student ID, a photo, or a zip code and birth date together. Laws such as the EU’s GDPR, and in the US FERPA for students’ education records, limit how it may be collected, stored, and shared, and research that collects it may need review by an institutional review board (IRB). Unlike a leaked password, leaked personal data can’t be changed or taken back. See Chapter 30, Chapter 34, and Chapter 8.

pip

Python’s built-in package manager, which installs packages from the Python Package Index (PyPI): pip install pandas. Run it inside an activated virtual environment, so packages land in your project and not in some other Python on your computer. See Chapter 14 and Chapter 15.

Programming language

A language for writing instructions a computer can carry out, like Python, R, JavaScript, Java, or C++. Most share the same big ideas (variables, loops, functions), so learning a second one is much easier than the first, even though the exact way you write things differs.

Provenance

The record of where a dataset came from and what has happened to it since: its source, when you got it, its license or terms of use, any limits on who may see it, and each step that changed it along the way. Write it down in a provenance.md beside the data, or in the README, the moment you download the file, because in three weeks you won’t remember. See Chapter 30 and Chapter 10.

REPL (Read–Eval–Print Loop)

An interactive prompt that reads a line of code, runs it, prints the result, and waits for the next one (the name spells out that loop). Type python with nothing after it in a terminal and you’re in Python’s REPL; type exit() to leave. A Jupyter kernel works the same way, which is why it remembers everything you’ve run. See Chapter 16.

Repository

A word with three meanings in research computing, and it helps to know which one someone means. A Git repository is a project folder whose whole history Git tracks, in a hidden .git folder inside it, and usually has a copy on a host such as GitHub. A data repository, such as Zenodo, OSF, or ICPSR, archives datasets and other research materials for the long term and can give each one a DOI so others can find and cite it, while an institutional repository is a university library’s archive of its own researchers’ papers, theses, and data, and often where a free open-access copy of a paper lives. See Chapter 31 and Chapter 25.

RLHF (Reinforcement Learning from Human Feedback)

One of the training steps that turns a raw text predictor into a helpful chat assistant, usually after it has first been trained on example conversations. People compare pairs of the model’s answers and pick the better one; those choices train a reward model, and the language model is then tuned to write answers that model scores highly. Much of that rating work is done by low-paid contract workers, sometimes reviewing disturbing material, which is one of the hidden costs behind AI tools. See Chapter 35, Chapter 37, and Wikipedia’s article.

Schema

The blueprint of a table or database: which columns it has, what type each one is, and what rules the values must follow (for example, “every order has a customer ID”). A schema is also a set of choices about the world. Once a gender column allows only two values, or a form offers only a fixed list of race categories, anyone who doesn’t fit gets squeezed in or left out, and the choice tends to stick because everything downstream depends on it. See Chapter 23, and Chapter 8 for the bigger picture.

Script

A file of code you run as a program from start to finish, like python analyze.py. Unlike a notebook, a script runs the same way every time, top to bottom, which makes it the better choice for anything you’ll repeat or share. See Chapter 17.

Sync client

A program, such as OneDrive, iCloud Drive, Google Drive for desktop, or Dropbox, that keeps a folder on your computer matched to a copy in the cloud and on your other devices. Syncing isn’t a backup: if you delete a file or save a bad edit, that syncs everywhere too. See Section 10.7.

Terminal

The application window you type commands into, such as Terminal on macOS or Windows Terminal. People often say “terminal” when they mean the shell, but they’re two things: the terminal is the window, and the shell (zsh, bash, or PowerShell) is the program inside it that reads and runs what you type. See Chapter 11.

Text editor

A program for writing and editing plain-text files, such as code, data, and configuration. Unlike a word processor, it saves exactly the characters you type and nothing else, which is what code needs. VS Code is the common all-purpose choice; nano and vim run inside a terminal, so they work on any server. See Chapter 12.

Traceback

The block of text Python prints when your code crashes. It lists the chain of calls that led to the error, from your code down to the line that failed, and ends with the error’s type and message. Start reading at the last line, which says what went wrong, then look up the list for the last line that’s in your code. See Chapter 7.

Version

A label, usually something like 2.2.3, that marks one specific release of a package or of Python itself. Many packages follow semantic versioning, where a change in the first number warns you that code written for the old version might break. Writing down the exact versions a project uses (pinning them) is what lets it run the same way later. See Chapter 14.

Virtual environment

A private Python setup for one project, with its own folder of installed packages, so what you install for one project can’t break another. python -m venv .venv creates one. You then activate it, with source .venv/bin/activate on macOS or Linux or .venv\Scripts\Activate.ps1 in Windows PowerShell, and your prompt starts with (.venv) while it’s on. See Chapter 15.

YAML

A plain-text format for settings files, written as key: value lines with indentation to show what belongs inside what. The indentation has to be spaces, never tabs, and a stray colon or tab is behind most YAML errors. Quarto, GitHub Actions, and conda’s environment.yml all use it. See Chapter 4.