Pandas Cookbook, Third Edition (William Ayd, Matthew Harrison)（Z-Library）

【One-Line Pitch】 A recipe-driven guide to doing real data work in pandas 2.x, written by a core maintainer and a veteran Python trainer. Best for Python users who already know the language and want fast, idiomatic answers for wrangling, reshaping, and scaling tabular data. 【Book Arc】 - **Opening (~0%–10%)**: Orients you to the book's philosophy and provenance — a foreword from pandas' creator, author credentials, and the premise that pandas is the de facto API for structured data even beyond "small data." Solves the "why this book, why now" question. - **Early (~10%–30%)**: Sets up the working environment and the recipe format itself: pandas 2.0+, Python 3.9+, NumPy, PyArrow, Jupyter notebooks, plus the "How to do it / How it works / There's more" structure and roughly two dozen shared datasets. Solves onboarding friction before any analysis begins. - **Early–Middle (~30%–50%)**: Foundations — constructing `Series`, `DataFrame`, and `Index` objects, customizing row/column labels, and inspecting core attributes like `dtype`, `name`, `index`, and `shape`. Solves the "what am I actually holding" problem. - **Middle (~50%–75%)**: The working core: selection and assignment, the pandas type system, the I/O system for reading/writing many formats, applying algorithms across structures, visualization with pandas and seaborn, reshaping, and group-by summarization. Solves day-to-day wrangling and aggregation. - **Late (~75%–90%)**: Temporal data types and time-series algorithms, then general usage and performance tips covering common pitfalls and their idiomatic fixes. Solves the "correct but slow" trap the authors explicitly warn about. - **Ending (~90%–100%)**: The pandas ecosystem — libraries that integrate with, extend, or complement pandas. Solves the question of where to go when pandas alone isn't enough. 【Key Takeaways】 - **Idiomatic pandas is a performance decision, not a style preference** (Early): the authors note order-of-magnitude differences between two working solutions to the same problem, so the book's real value is teaching the fast path, not just a path. - **`Series`, `DataFrame`, and `Index` are the three primitives everything else builds on** (Early–Middle): `Index` has no clean analogy in spreadsheets or SQL, yet it drives selection, joining, and alignment throughout the book. - **Explicit dtypes save memory and integrate better with typed systems** (Middle): choosing `int8` over an inferred 64-bit integer is framed as a deliberate developer choice, and the type system gets its own chapter because it keeps evolving. - **The recipe format is the pedagogy** (Early): each entry follows "How to do it," "How it works," and "There's more," so you can copy a solution and then understand why it works — useful for reference as much as linear reading. - **Real datasets, not toy examples, are the point** (Early): the book positions itself against the official docs' trivial fake data, using roughly two dozen documented datasets with provenance notes. - **Time series is treated as a first-class capability** (Late): date/time types and temporal algorithms get dedicated coverage, reflecting what pandas is historically famous for. - **Common pitfalls have known idiomatic fixes** (Late): the performance chapter exists because correct-looking pandas code is frequently inefficient, and the book names those traps directly. - **pandas is a hub, not an island** (Ending): NumPy, PyArrow, seaborn, and databases are woven throughout, and the final chapter maps the wider ecosystem — with the caveat that pandas itself stays in-memory. 【Reading Tips】 - **Skim the front matter, deep-read from Chapter 1 onward.** The preface, conventions, and dataset descriptions are setup; the substance starts with foundations and selection. - **Run the notebooks.** The authors explicitly recommend executing each recipe step and exploring variations rather than reading passively — this is a cookbook, not a novel. - **Keep the official pandas docs open alongside it.** The book assumes you'll cross-reference; the docs cover breadth with toy data, this book covers realistic data with depth. - **Treat the performance and ecosystem chapters as your exit ramp.** If you're already comfortable with wrangling, start at the performance tips and work backward only where needed. - **Watch the version boundary.** Examples target pandas 2.0+ and Python 3.9+, with PyArrow as an optional but frequently used dependency — older environments will hit friction. 【Coverage Limits】 The excerpts cover front matter, chapter listings, and the opening foundations material in detail, but later chapters (I/O, visualization, reshaping, group-by, time series, performance, ecosystem) are known only by their titles and brief descriptions. Specific recipes, benchmarks, and code patterns from those chapters are not represented here.