Pandas · Function

pd.read_pickle

Loads a pickled pandas object—or any other pickled object—from a file, URL, or binary stream. It can infer or apply compression and accepts storage-specific connection options.

How it works

The function opens a pickle source supplied as a path, URL, or binary file-like object, optionally decompresses it, and returns the object reconstructed from the serialized data. Compression can be inferred from supported filename suffixes or selected explicitly, while storage options configure remote access.

Syntax

pandas.read_pickle(filepath_or_buffer, compression='infer', storage_options=None)

Parameters

Name	Type	Description
`filepath_or_buffer`	`str, path object, or file-like object`	String, path-like object, or binary file-like object containing the pickle; URLs are also accepted.
`compression`	`str or dict, default 'infer'`	Compression mode or mapping for on-the-fly decompression; defaults to inferring compression from a path suffix.
`storage_options`	`dict, optional`	Additional options for the selected storage connection.

Returns

object — the unpickled pandas object or any other object stored in the file.

Key points

Loads a serialized pandas object or another Python object and returns the reconstructed value.
The source may be a string, path-like object, or binary file-like object.
URLs are valid sources, including remote object-storage locations.
The compression parameter enables decompression while reading stored data.
The default inference mode identifies compression from supported path suffixes.
storage_options passes backend-specific connection settings.
The function returns the object that was serialized in the source.

Common mistakes

Do not load pickle data from untrusted sources because unpickling can be unsafe.
A ZIP or TAR input must contain exactly one data file for this reader to select.

Example: Round-trip a DataFrame

import pandas as pd

frame = pd.DataFrame({"score": [10, 20]})
frame.to_pickle("scores.pkl")
restored = pd.read_pickle("scores.pkl")
print(restored)

Serializes a DataFrame, then loads it back from the pickle file.

When to use it

Restore locally serialized pandas data

Persisting a DataFrame between Python processes or runs.

Load compressed snapshots

Reading compressed pickle files without manually opening a decompressor.

Read from remote storage

Accessing pickle data through supported remote storage URLs.

Interview questions

QWhat kinds of inputs and outputs does read_pickle support?

It accepts a string, path-like object, or binary file-like object, and URLs are supported as well. The return value is the object stored in the pickle, not necessarily a DataFrame.

Follow-up: How would you load a compressed remote object? Pass the URL, select or infer compression, and provide any connection-specific settings through storage_options.

QHow does compression handling work?

With compression='infer', pandas detects compression from supported suffixes when the input is path-like. It supports gzip, bz2, zip, xz, zstd, tar, and several tar-compressed suffixes; use None to disable decompression.

Follow-up: What constraint applies to archive inputs? A ZIP or TAR archive must contain only one data file.

QWhen would you use storage_options?

storage_options is a dictionary for connection-specific settings such as host, port, username, or password. The options are used by the relevant URL or storage backend when opening the source.

Follow-up: Why should storage_options not be treated as a pickle-security mechanism? It configures access, but it does not make untrusted serialized data safe to unpickle.

pandas.DataFrame.to_pickle
pandas.Series.to_pickle
pandas.read_csv

Source: pandas.read_pickle. Checked 2026-10-15.