Pandas · Function

pd.read_feather

Loads a Feather-format file or file-like object into a pandas DataFrame. It supports column selection, threaded reading, storage options, and configurable dtype backends.

How it works

Pass a filesystem path, URL, or binary file-like object to load Feather data into a DataFrame. You can restrict columns, enable or disable threaded reading, provide connection options for remote locations, and choose the resulting dtype backend.

Syntax

pandas.read_feather(path, columns=None, use_threads=True, storage_options=None, dtype_backend=<no_default>)

Parameters

Name	Type	Description
`path`	`str, path object, or file-like object`	String, path-like object, or binary file-like object; URLs may use documented remote schemes.
`columns`	`sequence, default None`	Columns to load; when omitted, every column is read.
`use_threads`	`bool, default True`	Whether to parallelize reading with multiple threads.
`storage_options`	`dict, optional`	Additional options for the relevant storage connection.
`dtype_backend`	`{'numpy_nullable', 'pyarrow'}, default no_default`	Result dtype backend; documented choices are numpy_nullable and pyarrow.

Returns

DataFrame object stored in the file.

Key points

The function deserializes Feather data into a pandas DataFrame.
A string path may identify either a local resource or a remote URL.
The documented URL schemes include HTTP, FTP, S3, GS, and file URLs.
use_threads enables parallel reading and defaults to True.
storage_options carries connection-specific settings such as credentials or network parameters.
dtype_backend determines the dtype backend used by the returned DataFrame.

Common mistakes

Install and make pyarrow available before calling the function.
Leaving columns as None loads the complete file; provide a sequence to limit the read.
dtype_backend is experimental, so treat backend selection as an evolving interface.

Example: Read selected Feather columns

import pandas as pd

# Requires pyarrow and a Feather file at this path.
df = pd.read_feather("data.feather", columns=["user_id", "score"])
print(df)

Reads only the selected columns from a Feather file into a pandas DataFrame.

When to use it

Read a column subset

A Feather file contains many fields but an analysis needs only a few.

Load remote Feather data

A dataset is stored in an object store or accessible through a URL.

Choose the dtype backend

A pipeline requires a particular nullable or PyArrow dtype representation.

Interview questions

QWhat does pandas.read_feather return and what can path accept?

It loads a Feather-format object and returns the stored data as a DataFrame. The path may be a string, path-like object, or binary file-like object.

Follow-up: What remote locations can path represent? It can be a URL using schemes such as http, ftp, s3, gs, or file, with the documented file-URL host requirement.

QHow do columns and use_threads affect reading?

columns controls projection: when omitted, every column is read; when supplied, only the requested sequence is read. use_threads controls whether reading is parallelized with multiple threads and defaults to True.

Follow-up: Why pass columns in a large-file workflow? Selecting only needed columns can reduce the data that must be read and materialized.

QWhat is dtype_backend used for?

dtype_backend selects the dtype representation for the resulting DataFrame. The documented choices are numpy_nullable and pyarrow; without it, pandas does not use nullable data types by default.

Follow-up: What should be considered before choosing pyarrow? The option changes the resulting dtype backend and is marked experimental in the documentation.

pandas.DataFrame.to_feather
pandas.read_parquet
pandas.read_csv

Source: pandas.read_feather — pandas documentation. Checked 2026-10-15.