pandas: Python Data Analysis Toolkit

What is pandas?

pandas is a Python library for tabular and time series data, built around two objects: the Series (one labeled column) and the DataFrame (a full table with labeled rows and columns). It reads CSV, Excel, SQL, and HDF5 sources straight into a DataFrame, then lets you filter, group, reshape, and merge that data with a small set of methods instead of nested loops. NumPy supplies the array machinery underneath.

Core features

• ✓Labeled data structures — Series and DataFrame carry row and column labels, so pandas DataFrame operations align data by label instead of raw position.
• ✓Missing-data handling — NaN, NA, and NaT values are handled consistently across floating-point and non-floating-point columns.
• ✓groupby split-apply-combine — group by operations for aggregating or transforming data without writing manual loops.
• ✓Flexible merging and joining — combine two datasets the way a database JOIN would, keyed on an index or a column.
• ✓Reshaping and pivoting — turn long data wide, or back, with pivot_table and related methods.
• ✓Hierarchical indexing (MultiIndex) — attach more than one label per row or column axis.
• ✓Broad I/O support — csv reading plus Excel, SQL database, and HDF5 read/write in the same API.
• ✓Time series analysis tools — date range generation, frequency conversion, and moving-window statistics built in.

Installing pandas

pandas installs the same way most Python packages do. Via pip: `pip install pandas`. Via conda: `conda install -c conda-forge pandas`. Both pull in pandas plus its required dependencies — NumPy, python-dateutil, and, on Windows or Emscripten, tzdata — from PyPI or conda-forge respectively. Building from source is more involved: you need Cython in addition to the normal dependencies, then run `pip install .` from inside the cloned `pandas` directory, or `python -m pip install -ve . --no-build-isolation --config-settings editable-verbose=true` for an editable, development install.

Working with DataFrames

The GitHub README doesn't walk through a worked example — it lists the operations and links out to pandas' own user guide for each one. In practice, working with a DataFrame comes down to three habits: load data with a `read_*` function (`read_csv`, `read_excel`, `read_sql`), refer to columns and rows by label instead of position, and reach for `groupby()`, `merge()`, or `pivot_table()` instead of a loop. Step-by-step usage beyond that isn't clearly documented in the repo itself — the linked user-guide pages on missing data, groupby, merging, reshaping, and indexing carry the actual worked examples.

Alternatives to consider

Frequently asked questions

The problem it solves

Raw tabular data in Python — a CSV read into a list of dicts, or a NumPy array with no column names — has no memory of what its rows and columns mean. Every operation, filtering by date or summing by category, has to be rebuilt from scratch with index math or manual loops each time. pandas attaches labels to both axes so "rows where date > X" or "sum revenue by region" become one method call, and it folds CSV, Excel, SQL, and HDF5 input into the same DataFrame shape so the source format stops being the analysis's problem.

Best use cases

• •Cleaning a messy CSV or Excel export before it goes into a report or a model.
• •Aggregating transaction or log data by day, category, or user with groupby.
• •Joining two datasets — orders and customers, say — on a shared key.
• •Running time series analysis on a metric: resampling to weekly totals, computing rolling averages.

Who should try it — and who should skip

Anyone doing exploratory data analysis, ETL scripts, or reporting in Python is the target user — the CSV-to-chart-to-report loop is what pandas was built for. Skip it, or at least don't start there, if the data already lives in a warehouse and a SQL query would answer the question faster, or if the dataset is too large for one machine's RAM and you'd be reaching for chunked workarounds from day one.

Related repositories