What Is Pandas? Python's Data Analysis Toolkit

If you’ve ever opened a CSV in Excel, sorted it, filtered it, made a pivot table, and graphed a column — pandas is the Python tool that does all of that programmatically, on datasets Excel would refuse to open. It is the foundation of the Python data ecosystem and the daily driver for data analysts, scientists, and ML engineers.

This post explains what pandas is, the ideas behind it, and walks through your first end-to-end script.

What pandas actually is

Pandas is a Python library for working with tabular data — anything that fits naturally into rows and columns. It was created in 2008 by Wes McKinney at a hedge fund that needed a better way to handle financial time series in Python.

Three things to internalise:

• It is built on top of NumPy, which gives it fast, vectorised numerical operations.
• Its API is heavily inspired by R’s data.frame and SQL, with Python conveniences.
• It is the default Python tool for cleaning, transforming, and exploring tabular data before you do anything else (model it, visualise it, ship it).

If your data has rows and columns, you probably want pandas.

Two core data structures

Pandas has two objects you’ll use constantly.

Series

A Series is a one-dimensional labelled array. Think of it as a single column of a spreadsheet.

import pandas as pd

prices = pd.Series([10.5, 12.0, 9.75, 14.25])
print(prices)
# output:
# 0    10.50
# 1    12.00
# 2     9.75
# 3    14.25
# dtype: float64

The numbers on the left are the index. By default it’s a range; you can replace it with anything meaningful:

prices = pd.Series([10.5, 12.0, 9.75], index=["apple", "banana", "cherry"])
print(prices["banana"])
# output: 12.0

A Series carries a single data type for all its values — floats here, but it could be ints, strings, dates, or booleans.

DataFrame

A DataFrame is a two-dimensional labelled table — rows and columns, like a spreadsheet or a SQL table.

import pandas as pd

data = {
    "product": ["apple", "banana", "cherry"],
    "price":   [10.5, 12.0, 9.75],
    "stock":   [40, 12, 200],
}
df = pd.DataFrame(data)
print(df)
# output:
#   product  price  stock
# 0   apple  10.50     40
# 1  banana  12.00     12
# 2  cherry   9.75    200

A DataFrame is a collection of Series — each column is a Series, and they all share the same index. That mental model — “rows are aligned columns” — is more accurate than thinking of it as a list of rows.

Every column can have a different data type, which is exactly what real-world data needs.

Why so much industry runs on pandas

A short list of why pandas is everywhere in data work:

• Loading anything. read_csv, read_json, read_excel, read_parquet, read_sql — pandas reads the formats data actually lives in.
• Vectorised speed. Operations on whole columns are compiled C under the hood; a million-row sum takes milliseconds.
• Expressive selection and filtering. df[df["price"] > 10] reads like SQL but is just Python.
• Group-by and aggregation. Pivot tables, group means, rolling windows — one line each.
• Time series tools. Calendar-aware resampling, time zones, business days.
• Plays nice with the ecosystem. NumPy under it, scikit-learn next to it, Matplotlib and Seaborn for charts, Jupyter for notebooks.
• Ubiquity. Every Python data tutorial, course, and Stack Overflow answer uses it. That network effect compounds.

When a Python data role lists “pandas” in the requirements — which is most of them — this is what they mean.

Installing pandas

Pandas is on PyPI. Inside your project’s virtual environment:

pip install pandas

That pulls in NumPy as a dependency. For reading Excel files you may also want:

pip install openpyxl

For Parquet:

pip install pyarrow

If you’re using a notebook environment (Jupyter, Google Colab), pandas is usually pre-installed.

Verify with:

import pandas as pd
print(pd.__version__)
# output: 2.2.2   (or similar)

pd is the universal alias. Use it; everyone else does.

A first end-to-end example

Imagine a CSV file sales.csv:

date,product,units,revenue
2026-01-04,apple,12,18.00
2026-01-04,banana,7,8.40
2026-01-05,apple,15,22.50
2026-01-05,cherry,3,11.25
2026-01-06,banana,11,13.20

A short script that loads it and explores:

import pandas as pd

df = pd.read_csv("sales.csv")

print(df.head())
# output:
#         date product  units  revenue
# 0 2026-01-04   apple     12    18.00
# 1 2026-01-04  banana      7     8.40
# 2 2026-01-05   apple     15    22.50
# 3 2026-01-05  cherry      3    11.25
# 4 2026-01-06  banana     11    13.20

head() shows the first five rows — a quick sanity check that the data loaded as expected.

For a summary of numeric columns:

print(df.describe())
# output:
#            units    revenue
# count   5.000000   5.000000
# mean    9.600000  14.670000
# std     4.722288   5.789948
# min     3.000000   8.400000
# 25%     7.000000  11.250000
# 50%    11.000000  13.200000
# 75%    12.000000  18.000000
# max    15.000000  22.500000

describe() gives you count, mean, standard deviation, min, max, and quartiles for every numeric column — a one-line overview of the shape of your data.

For the data types and missing-value count:

print(df.info())
# output:
# <class 'pandas.core.frame.DataFrame'>
# RangeIndex: 5 entries, 0 to 4
# Data columns (total 4 columns):
#  #   Column   Non-Null Count  Dtype
# ---  ------   --------------  -----
#  0   date     5 non-null      object
#  1   product  5 non-null      object
#  2   units    5 non-null      int64
#  3   revenue  5 non-null      float64
# dtypes: float64(1), int64(1), object(2)
# memory usage: 288.0 bytes

head, describe, and info are the first three things you run on any new dataset. They tell you what you’re working with before you write any analysis.

A first transformation

Beyond looking, a tiny taste of what pandas lets you do.

Total revenue per product:

totals = df.groupby("product")["revenue"].sum()
print(totals)
# output:
# product
# apple     40.50
# banana    21.60
# cherry    11.25
# Name: revenue, dtype: float64

That single line is a group-by and a sum. The equivalent in pure Python with dictionaries and for loops is a dozen lines and easy to get wrong.

Filtering rows:

big_sales = df[df["units"] > 10]
print(big_sales)
# output:
#         date product  units  revenue
# 0 2026-01-04   apple     12    18.00
# 2 2026-01-05   apple     15    22.50
# 4 2026-01-06  banana     11    13.20

Adding a derived column:

df["price_per_unit"] = df["revenue"] / df["units"]
print(df.head())
# output:
#         date product  units  revenue  price_per_unit
# 0 2026-01-04   apple     12    18.00            1.50
# 1 2026-01-04  banana      7     8.40            1.20
# 2 2026-01-05   apple     15    22.50            1.50
# 3 2026-01-05  cherry      3    11.25            3.75
# 4 2026-01-06  banana     11    13.20            1.20

This is the style of pandas code that fills most data jobs: load, look, filter, group, derive, repeat.

Where pandas fits in the stack

A typical Python data workflow:

• pandas for ingestion, cleaning, exploration, feature engineering.
• NumPy for the numerical operations underneath.
• Matplotlib or Seaborn for plotting.
• scikit-learn when you want a model — see What Is Machine Learning?.
• Jupyter as the interactive environment that ties it together.

In production data pipelines, pandas often hands off to:

• DuckDB or SQL for very large data.
• Polars (a faster, Rust-based DataFrame library with a similar API) when pandas becomes a bottleneck.
• PySpark for genuinely huge datasets.

Pandas remains the right starting point. You can always migrate later; you almost never need to.

What pandas is not

• Not a database. It loads data into memory. Datasets larger than your RAM need DuckDB, Spark, or chunked reading.
• Not a stats library. It has basic descriptive stats; for tests and models, use SciPy or statsmodels.
• Not the fastest DataFrame library anymore. Polars and DuckDB beat it on many benchmarks. But pandas is the most universally known.

Recap

You now know:

• Pandas is Python’s standard library for tabular data
• A Series is a labelled 1D array; a DataFrame is a labelled 2D table
• It’s the daily tool for analysts, scientists, and ML engineers
• pip install pandas, then import pandas as pd
• read_csv, head, info, and describe are your first four moves on any dataset
• Pandas sits on NumPy and hands off to plotting, ML, and SQL tools

Next steps

The next post goes deeper into the DataFrame — selecting columns, indexing rows with loc and iloc, filtering, and sorting.

→ Next: Pandas DataFrames: Reading, Selecting, and Filtering

Related: What Is Python?, Python Dictionaries, What Is Machine Learning?.

Questions or feedback? Email codeloomdevv@gmail.com.