Data Analysis with Python Cheat Sheet

Updated 2026-07-17

Next Topic: Data Science Core Cheat Sheet

🎯Take a practice test on this topic7 practice tests · 241 questions→

🧠Study flashcards on this topic151 cards · spaced repetition→

Data analysis with Python centers on Pandas for tabular data manipulation and NumPy for numerical computing. Pandas provides DataFrames (2D labeled data structures) enabling SQL-like operations, while NumPy delivers vectorized array computations orders of magnitude faster than pure Python. Pandas 3.0 introduced Copy-on-Write by default, a dedicated str dtype, and pd.col() expressions, making data manipulation more predictable and performant. Together with NumPy's modern random Generator API (default_rng), they form the foundation of Python's data science ecosystem, handling everything from cleaning messy datasets to time series analysis and statistical aggregation.

What This Cheat Sheet Covers

This topic spans 23 focused tables and 195 indexed concepts, 151 flashcards, 7 practice tests with 241 questions. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Reading and Writing DataTable 2: DataFrame Creation and InspectionTable 3: Indexing and SelectionTable 4: NumPy Array CreationTable 5: NumPy Indexing and SlicingTable 6: Data Cleaning and Missing ValuesTable 7: Data Transformation and ManipulationTable 8: NumPy Array ManipulationTable 9: Aggregation and GroupingTable 10: Data Binning and EncodingTable 11: Combining DataFramesTable 12: Sorting and RankingTable 13: String OperationsTable 14: Time Series and DatetimeTable 15: NumPy Mathematical OperationsTable 16: NumPy Statistical FunctionsTable 17: NumPy Broadcasting and VectorizationTable 18: NumPy Linear AlgebraTable 19: NumPy Random SamplingTable 20: Data Type OperationsTable 21: Advanced Indexing TechniquesTable 22: Performance OptimizationTable 23: Visualization Integration

Quick IndexSubscribe to unlock

A jump-to index of every table row in this cheat sheet.

Mind MapSubscribe to unlock

An interactive map of every table and concept in this topic.

Table 1: Reading and Writing Data

Every analysis starts by getting data into a DataFrame and ends by writing results back out, and Pandas reads almost any format you'll meet. CSV and Excel cover the everyday cases, but they're untyped text, so Pandas has to guess every column's type on each read. Columnar formats like Parquet, Feather, and ORC store the schema next to the data and compress it, so a round trip hands back the dtypes you started with.

Method	Example	Description
read_csv()	`df = pd.read_csv('data.csv')`	• Reads a comma-separated values file into a DataFrame • infers each column's type from the values, so pass `dtype={'zip': 'str'}` to stop digits with leading zeros parsing as `int64`.
read_excel()	`df = pd.read_excel('data.xlsx', sheet_name='Sheet1')`	• Reads Excel files (.xls, .xlsx, .xlsm, .xlsb) and OpenDocument (.odf, .ods, .odt) • `sheet_name=None` returns a dict of every sheet.
to_csv()	`df.to_csv('output.csv', index=False)`	• Exports a DataFrame to CSV • writes the index as an extra unnamed column unless you pass `index=False`.
to_excel()	`df.to_excel('output.xlsx', sheet_name='Data')`	• Exports to an Excel file via the `openpyxl` or `xlsxwriter` engine • a plain call rewrites the whole file, so use `pd.ExcelWriter` to put several sheets in one workbook.
read_json()	`df = pd.read_json('data.json', orient='records')`	• Parses JSON into a DataFrame • `orient` names the layout to expect (`records`, `columns`, `split`, `table`), since JSON can hold the same table many ways.
read_sql()	`df = pd.read_sql('SELECT * FROM t', conn)`	• Runs a SQL query or reads a table into a DataFrame • takes an ADBC connection, a SQLAlchemy connectable or URL string, or a sqlite3 connection • Pandas doesn't sanitize SQL, so pass values through `params`.
read_parquet()	`df = pd.read_parquet('data.parquet')`	• Reads columnar Parquet files • the schema travels with the data, so every Pandas dtype survives the round trip.
to_parquet()	`df.to_parquet('data.parquet', index=False)`	• Exports to Parquet, `snappy`-compressed by default • `partition_cols=['year']` writes a directory tree of `year=...` folders instead of a single file.
read_feather()	`df = pd.read_feather('data.feather')`	• Reads the Feather binary columnar format • preserves the schema, which suits fast handoffs between Python and R and intermediate pipeline results.
read_orc()	`df = pd.read_orc('data.orc')`	• Reads ORC (Optimized Row Columnar) files through `pyarrow` • common in Hadoop and Hive setups, and not supported on Windows yet.
read_html()	`tables = pd.read_html(url)`	• Extracts every HTML table on a page into a list of DataFrames, never a bare one • needs `lxml`, or `bs4` plus `html5lib`.

Table 2: DataFrame Creation and Inspection

The first thing you do with a new DataFrame is look at it, and these are the methods you'll type before anything else. head, info, and describe give you a fast read on structure, types, and summary stats, while shape, dtypes, and nunique answer the "how big and how varied" questions that shape your next move. Watch what each one quietly leaves out: info counts non-null values rather than missing ones, and describe skips text columns unless you ask for them.

Method	Example	Description
DataFrame()	`df = pd.DataFrame({'A': [1, 2], 'B': [3, 4]})`	• Creates DataFrame from dict of lists • keys become column names.

The rest of this section is part of the full cheat sheet

Table 3: Indexing and Selection

Pulling out exactly the rows and columns you want is a daily task, and the choice of tool matters. The core split is loc (by label, inclusive) versus iloc (by integer position, exclusive), with Boolean indexing and query handling conditional filters and isin/between covering membership and ranges. The new pd.col() expression replaces lambdas for cleaner, bug-resistant column references.

Technique	Example	Description
loc[]	`df.loc[0:5, 'A':'C']`	• Selects by label (row/column names) • inclusive of end boundary.

The rest of this section is part of the full cheat sheet

Table 4: NumPy Array Creation

NumPy work starts by building an array, and the call you pick fixes its shape, its dtype, and whatever is already sitting in it. array() wraps a Python list, zeros, ones, and full pre-fill a grid you plan to overwrite, and arange and linspace generate evenly spaced runs of numbers. The split worth remembering: arange takes a step size and stops short of the end, while linspace takes a count and lands on it.

Function	Example	Description
array()	`arr = np.array([1, 2, 3])`	• Creates array from a Python list or tuple • infers one shared dtype, promoting ints to float if any value is a float.

The rest of this section is part of the full cheat sheet

Table 1: Reading and Writing Data

Method	Example	Description
read_csv()	`df = pd.read_csv('data.csv')`	• Reads a comma-separated values file into a DataFrame • infers each column's type from the values, so pass `dtype={'zip': 'str'}` to stop digits with leading zeros parsing as `int64`.
read_excel()	`df = pd.read_excel('data.xlsx', sheet_name='Sheet1')`	• Reads Excel files (.xls, .xlsx, .xlsm, .xlsb) and OpenDocument (.odf, .ods, .odt) • `sheet_name=None` returns a dict of every sheet.
to_csv()	`df.to_csv('output.csv', index=False)`	• Exports a DataFrame to CSV • writes the index as an extra unnamed column unless you pass `index=False`.
to_excel()	`df.to_excel('output.xlsx', sheet_name='Data')`	• Exports to an Excel file via the `openpyxl` or `xlsxwriter` engine • a plain call rewrites the whole file, so use `pd.ExcelWriter` to put several sheets in one workbook.
read_json()	`df = pd.read_json('data.json', orient='records')`	• Parses JSON into a DataFrame • `orient` names the layout to expect (`records`, `columns`, `split`, `table`), since JSON can hold the same table many ways.
read_sql()	`df = pd.read_sql('SELECT * FROM t', conn)`	• Runs a SQL query or reads a table into a DataFrame • takes an ADBC connection, a SQLAlchemy connectable or URL string, or a sqlite3 connection • Pandas doesn't sanitize SQL, so pass values through `params`.
read_parquet()	`df = pd.read_parquet('data.parquet')`	• Reads columnar Parquet files • the schema travels with the data, so every Pandas dtype survives the round trip.
to_parquet()	`df.to_parquet('data.parquet', index=False)`	• Exports to Parquet, `snappy`-compressed by default • `partition_cols=['year']` writes a directory tree of `year=...` folders instead of a single file.
read_feather()	`df = pd.read_feather('data.feather')`	• Reads the Feather binary columnar format • preserves the schema, which suits fast handoffs between Python and R and intermediate pipeline results.
read_orc()	`df = pd.read_orc('data.orc')`	• Reads ORC (Optimized Row Columnar) files through `pyarrow` • common in Hadoop and Hive setups, and not supported on Windows yet.
read_html()	`tables = pd.read_html(url)`	• Extracts every HTML table on a page into a list of DataFrames, never a bare one • needs `lxml`, or `bs4` plus `html5lib`.

Table 2: DataFrame Creation and Inspection

Method	Example	Description
DataFrame()	`df = pd.DataFrame({'A': [1, 2], 'B': [3, 4]})`	• Creates DataFrame from dict of lists • keys become column names.

Table 3: Indexing and Selection

Technique	Example	Description
loc[]	`df.loc[0:5, 'A':'C']`	• Selects by label (row/column names) • inclusive of end boundary.

Table 4: NumPy Array Creation

Function	Example	Description
array()	`arr = np.array([1, 2, 3])`	• Creates array from a Python list or tuple • infers one shared dtype, promoting ints to float if any value is a float.