mrkeyoor.comAsk

pandas-dev/pandastested 11 Nov 2026commit f1cc1a1

pandas review

Use pandas when labeled tables and compatibility with existing Python analysis tools meet your needs.

  • README snippet failed
  • 101.3 MB on disk
Timed outinstall
  • 3 dependencies
Dataevaluationupdated 22 Nov 2026
+52stars / 7d

pandas is a Python library for cleaning, joining and analyzing tables and time-series data. Its DataFrame and Series structures support repeatable analysis of files and database results.

Verdict

Use pandas when labeled tables and compatibility with existing Python analysis tools meet your needs. Avoid choosing it as the default for slow native table operations when Polars fits the downstream workflow; Jodhan's comparison supports evaluating that alternative. Our November 22 installation obtained pandas 3.1.1 in 1.6 seconds and stopped at README example discovery. The November 11 source installation timed out at 900 seconds.

pandas infographic: analyze tables and time-series data in Python, plus its uses, our test result, pitfalls and who it suits
Tested 22 November 2026: the README snippet failed. Before that: 11 Nov: source install timed out; uv install took 1.6 s.

We ran it

Lab card: what happened when we ran pandas
Lab card from our sandbox run
Screenshot of pandas (pandas.pydata.org)
pandas.pydata.org
Install✗ timed out · 900s
Build—
Repo2696 files~770,437 lines of source · 62.4 MB · 14 CI workflows

Answers from our run

Does pandas build from source?

The dependency install failed, and the project has no separate build step. We cloned commit f1cc1a1 into a clean Debian container with 3 CPUs and no project-specific setup.

Who should not use pandas?

Teams whose main requirement is faster native table operations and whose downstream tools accept Polars results.

What are the alternatives to pandas?

Polars, NumPy, DuckDB. Use pandas when labeled tables and compatibility with existing Python analysis tools meet your needs.

Setup3/5In our 2026-11-22 run it installed in 1.6 s, but installing from source timed out in our sandbox.Editor’s note: The package install completed, example discovery failed, and a separate source install timed out; these outcomes do not assess setup across supported environments.
Docs3/5The 6,233-character README has no usage instructions or examples of its own and points to separate documentation, which we did not assess.Editor’s note: The User Guide supplies a beginner route and a complete construction example; the completed run covers README example discovery rather than full API documentation quality.
Community5/5Last push on 2026-11-22, 50,595 stars and maintainers answered or closed 10 of the 10 most-discussed issues we sampled.Editor’s note: Sampled issues contain maintainer recommendations and individual resolutions, without representative support-response or adoption measurements.
Maturity4/5Latest release v3.1.1 on 2026-11-04 and repository created 2010-08-24.Editor’s note: Release metadata and issue discussions document maintenance activity, while the completed run contains no application-correctness or production-reliability result.

How we score

Discussed on

  1. hnPandas 2.0325 points
  2. hnOVH forgot they donated documentation hosting to Pandas149 points
  3. hnPandas extension arrays95 points
  4. hnPDEP-13: The Pandas Logical Type System46 points
  5. hnPandas 3.0.0 Released5 points

Who it’s for

Analysts turning repeated spreadsheet preparation into Python code.
Developers whose analysis tools accept pandas DataFrames and whose current performance is acceptable.
Teams that define how labels and missing observations should behave in reports.

Who it’s NOT for

Teams whose main requirement is faster native table operations and whose downstream tools accept Polars results.
Anyone choosing pandas on the assumption that it is a complete predictive-analytics framework.
Contributors expecting our recorded source installation to finish within a 900-second allowance.

Setup reality

The README offers released packages through PyPI and conda-forge and directs contributors to separate source-installation instructions. Our November 22 package installation obtained pandas 3.1.1 in 1.6 seconds. Our November 11 source installation reached its 900-second timeout.

File-format and output methods can require optional packages. The installation guide says pandas raises ImportError when a method needs an optional dependency that is absent. The pandas 3.1.1 PyPI entry requires Python 3.11 or newer.

pandas suits Python analysis that needs labeled tables and compatibility with existing tools. Our November 22 run installed pandas 3.1.1 in 1.6 seconds, then stopped when the recipe found no Python example block in the README. For choosing an engine, the useful distinction comes from Dave Jodhan's comparison: native table operations favored Polars, while a custom Python transformation showed little improvement. Keep pandas when its interface and existing integrations meet your needs; consider Polars when those native operations are the expensive part of your work.

What people use pandas for

A TrustRadius reviewer uses pandas to calculate API response times and resource consumption in an analytics framework. That is a concrete operational-analysis workflow: service measurements become data to clean, summarize and inspect. The reviewer also describes statistical analysis, cleaning and charting as strengths, while reporting that some error messages provide little help with debugging. Their operational-analysis account supports these specific uses rather than a universal recommendation for every analytics task.

Who used it Task and outcome Practical use
A TrustRadius reviewer Calculated API response times and resource consumption. Operational metrics account Analyze service measurements.
Dave Jodhan Compared externally stored datasets and generated a six-million-row dataset for filtering, arithmetic, string operations and a custom Python transformation. Comparison methodology Identify which transformations justify changing engines.
MrKeyoor's sandbox Attempted installation of source checkout f1cc1a1 on November 11; installation timed out at 900 seconds. Assess contributor setup within a fixed time allowance.

Jodhan's comparison is useful because it separates operations instead of treating a library switch as one indivisible improvement. His generated dataset exercised filtering, arithmetic, string manipulation and a custom Python function. The custom transformation showed little benefit from Polars. His recommendation for existing pandas users also considers acceptable current performance and the libraries that consume the resulting DataFrame. That makes the next decision specific: identify the expensive operation and the required output type.

What you can do with pandas

The pandas README documents grouping, joining, reshaping, missing-data handling and time-series operations. These are building blocks for repeatable preparation: load a file, select relevant observations, combine related tables and calculate a report. Its feature list includes CSV, Excel and database input, along with date-range generation, frequency conversion, moving-window statistics, date shifting and lagging. The same library therefore covers both ordinary tables and observations whose dates affect the calculation.

Labels are part of that interface. The README describes automatic and explicit alignment, so pandas can match observations by labels during computations. That is useful when two inputs refer to the same entities in different orders. It also makes labels part of the calculation's meaning. Before writing a recurring report, decide which column or row labels identify an observation and how absent entries should contribute. A successful calculation still needs an expected result that reflects the report's rules.

What happened when we ran it

Our November 22 package run installed pandas 3.1.1 in 1.6 seconds. The subsequent snippet step failed because the README had no Python code block for the recipe to run. The installation result and snippet failure record two different steps: the package installation completed, and example discovery stopped. The recorded failure was finding executable documentation, rather than an exception from a pandas calculation.

The November 22 run configuration records checkout ef7d968a259a and a fresh unprivileged container with 2 CPUs, 4 GB RAM, no published ports and no secrets. The recipe installed pandas from PyPI rather than building that checkout. Its recorded image label was ghcr.io/astral-sh/uv:python3.12-bookworm. Our earlier source attempt used commit f1cc1a1 on November 11 with a configured 3 CPUs and 8 GB RAM and the same image label.

Question or step Recorded answer Verdict Why
Did cloning complete? Clone completed in 4 seconds. Completed The quickstart-lib recipe recorded a completed clone.
Did package installation complete? pandas 3.1.1 installed in 1.6 seconds using uv pip install pandas in a fresh virtual environment. Completed The step recorded the installed version and elapsed time.
How much installed disk space was recorded? 101.3 MB in the environment's site-packages directory. Measured The measurement covers that directory after installation.
How many other packages installed? 3 packages besides pandas installed. Measured The recipe counted packages installed alongside pandas.
Did the example step complete? FAILED: the README had no Python code block to run. Failed The recipe stopped at example discovery.
What finding did the recipe record? no-snippet: the README had no Python code block to run. Recorded This identifies the stopped step.
Did cleanup complete? Teardown: OK. Completed The teardown record reports success.
Did the November 11 source installation complete? Installation timed out at 900 seconds on f1cc1a1. Failed The harness reached its time allowance.

Our November 11 source measurement recorded 2,696 files, approximately 770,437 lines of source and a 62.4 MB checkout. Its repository scan found 14 CI workflow files, no Dockerfile and no tests directory. Those last items describe scanner results; the README links to the project's test workflow. The two installation paths answer different setup questions: November 22 obtained a released package, while November 11 attempted installation from source and reached its time limit.

Where people get stuck

Missing optional dependencies can raise ImportError when a pandas method needs another package, according to the installation guide. The guide gives pandas.read_hdf() and DataFrame.to_markdown() as examples with separate requirements. Choose dependencies around the methods your application calls. Installing the core package and installing every file-format or output dependency are separate tasks, which helps explain why an import can succeed before a particular reader or writer fails.

Problem Evidence-backed response Reported outcome
Windows reports that pip is not recognized Cameron Simpson suggested py -m pip install pandas. Python.org exchange Dave Curoe replied that it worked.
A method needs an optional package Install the relevant documented dependency or extra. Dependency guidance The guide explains the resulting ImportError.
Code repeatedly calls append to grow a table Collect values before construction, or collect frames before one concat. Append discussion #35407 The maintainer recommends batching construction.
A contributor cannot run pylint checks The contributor reinstalled pre-commit. Contributor issue #48855 The contributor reported that the command then ran.
Error messages provide few debugging clues The TrustRadius reviewer reports this difficulty. Error-handling account The account gives no concrete remedy.

The append discussion explains that index and value data must be copied to create the result. Its recommendation is to build a list of values and pass it to a constructor, or build a list of frames and perform one concat. That changes how incoming data is collected. A Python list can gather records before table construction; repeatedly creating a larger DataFrame is a different operation with different copying behavior. The distinction is more useful than replacing one method name throughout a loop.

Missing values can change the meaning of your answer

Issue #32265 asks whether np.nan and pd.NA could coexist in a floating-point Series and signify different things. The missing-value discussion separates technical representation from the decision about intended behavior. For a report, the important question is whether an absent observation and an invalid numerical result should contribute in the same way. Writing down those expectations gives a concrete basis for checking aggregates and transformations that encounter either case.

The historical all-NaN summation issue #9422 describes different outcomes depending on whether bottleneck was installed: pandas' internal implementation returned NaN, while the cited bottleneck implementation returned zero. That disagreement illustrates why an expected aggregate needs more detail than 'sum this column.' An empty or entirely missing group may mean zero in one report and unknown in another. The reporting rule should determine the expected output before implementation choices determine it accidentally.

Arrow changes involve packaging and correctness

The update to issue #54466 says pandas 3.0 would not require PyArrow and would use it by default, when installed, for the new string dtype. That dependency decision supersedes the issue's original proposal to make PyArrow mandatory. The release qualification belongs beside the claim: this is the recorded decision for pandas 3.0. An earlier proposal and the later update describe different stages of the same discussion.

The Arrow type-system discussion #61618 connects adoption to null semantics, specifically NA versus NaN. Storage representation and the meaning of values therefore belong in the same migration review. When an application changes column representations, compare expected values and missing-value behavior alongside its packaging requirements. A dtype choice determines how values are represented and handled; treating it solely as a storage change leaves part of the application contract unspecified.

pandas vs Polars, NumPy, SQL and Spark

Jodhan's comparison favored Polars for native operations, while his per-element Python transformation showed little improvement. His workload comparison supports choosing around the operation that consumes time and the tools that receive the result. Polars' own README describes a DataFrame API and SQL running through the same optimizer and engine. DuckDB's README describes an analytical database running inside an application, with direct file queries and Python integration.

Choice When to consider it Evidence behind the distinction
pandas Existing tools consume pandas DataFrames and current performance is acceptable. Jodhan's recommendation includes acceptable performance and surrounding libraries.
Polars The expensive work maps to native table operations. Jodhan reports operation-specific gains; the Polars README documents DataFrame and SQL queries.
NumPy The consumer needs numerical arrays. The pandas README identifies NumPy as an array dependency; the conversion guide explains conversion costs.
DuckDB / SQL SQL over CSV or Parquet expresses the task directly. The DuckDB README documents in-process execution and direct file queries.
Spark / pandas API on Spark Distributed processing is the requirement under consideration. Databricks' comparison discusses the pandas API on Spark.
A spreadsheet The workflow centers on a worksheet and direct graphical interaction. The official spreadsheet comparison maps worksheet concepts to pandas structures.

Conversion is part of the pandas-versus-NumPy decision. The basic-functionality guide says to_numpy() can copy data and coerce values when a Series or Index uses an ExtensionArray. Choose the representation required by the next consumer and account for that conversion in the pipeline. DuckDB offers a different expression of the work: its README shows CSV and Parquet files referenced directly in SQL's FROM clause, which can preserve an existing analytical query rather than translating it into successive DataFrame operations.

What the other reviews get wrong

Databricks' row-store explanation and JetBrains' unqualified single-threaded description need narrower readings. The Databricks comparison describes complete rows stored continuously in memory, while pandas' basic-functionality guide describes containers for underlying arrays. The JetBrains comparison calls pandas single-threaded, while the pandas acceleration documentation describes certain numerical and boolean operations accelerated through numexpr, which uses multiple cores. An operation and its execution path are more precise than either blanket description.

Pankaj Pandey's beginner guide promises big-data work without excessive resource consumption. That is too broad for an adoption recommendation: the TrustRadius account reports difficulties with large data, and resource requirements depend on the work being performed. Read the beginner guide's claim alongside the reviewer's experience. Neither a reassuring adjective nor one user's difficulty supplies a universal capacity figure for pandas.

The October review's pandas 3.0.6 label was a dated release snapshot. The November 4, 2026 PyPI record identifies pandas 3.1.1 and requires Python 3.11 or newer. Installation guidance listing Python 3.9, 3.10 and 3.11 belongs to pandas 2.1 rather than this package version. These corrections keep release identity and interpreter requirements together: use the requirement for the version being installed, rather than carrying a historical compatibility list into a current setup decision.

Questions people ask

pandas provides DataFrame and Series structures for labeled data analysis, including tables and time series. The README documents cleaning-related missing-data handling, grouping, joining and reshaping, alongside file and database input. For a first installation, it offers PyPI and conda-forge packages. For a first lesson, the User Guide recommends 10 minutes to pandas, followed by the data-structures introduction and essential basic functionality. Those routes answer different needs: obtaining the library and learning its table model.

What is pandas in Python, and what is it used for?

pandas is a Python library for working with labeled tables and time series. Its DataFrame represents a table, and its Series represents a labeled column. The README lists grouping, joining, reshaping and missing-data handling among its features. Use it when those operations belong in repeatable Python analysis, such as preparing recurring exports or calculating operational metrics. A DataFrame is one of pandas' structures, rather than a separate competing library.

Why is pip not recognized when I try to install pandas on Windows?

Dave Curoe resolved his reported Windows installation problem with py -m pip install pandas, following Cameron Simpson's suggestion. The Python.org discussion records the error and his successful response. The command invokes pip through the Python launcher rather than relying on a standalone pip command being found. It is the documented resolution for that exchange, which gives a beginner a concrete command tied to an identifiable error.

How do I open IPython after installing pandas?

The Python.org response recommends installing IPython separately with py -m pip install ipython. Curoe's follow-up question distinguishes the pandas library from the interactive shell used in his lecture. Installing the library and obtaining that shell are separate tasks. Follow the installation command for the shell itself rather than copying a lecturer's drive path, which describes a location on the lecturer's machine.

What does the name pandas stand for?

The name pandas comes from panel data, according to the creator of Learn Pandas in 1 hour!. The same introduction explains Series as a labeled column and DataFrame as a labeled table, then proceeds into importing and manipulating data. The name's origin is a short piece of background; the structures and operations explain what a new user will actually work with in Python.

What Python knowledge do I need before learning pandas?

The creator of Learn Pandas in 1 hour! recommends learning Python through lists before starting. That prerequisite connects directly to table construction: a beginner should understand a collection of values before learning how pandas labels and organizes it. The append discussion provides another reason lists matter, because collecting values before constructing a table is the recommended alternative to repeated append-based DataFrame growth.

Can I learn the basics of pandas in one hour?

Learn Pandas in 1 hour! introduces Series, DataFrames and importing the library as a starting course. Its introduction describes the goal as getting viewers started with pandas. Work through those structures before applying them to a recurring report. Learning how to select and transform a table is one task; defining the report's labels, absent observations and expected aggregates gives those operations their application-specific meaning.

Where should a beginner start with the pandas documentation?

The User Guide recommends 10 minutes to pandas for new users, followed by Intro to data structures and Essential basic functionality. Its documentation overview points readers to the API reference for individual methods. For browser-based lessons, the full-course creator uses Google Colab and also names Anaconda and VS Code as options. Choose the lesson sequence first, then use a notebook or editor that fits your learning environment.

How do I create a pandas DataFrame?

The User Guide creates a DataFrame with pd.DataFrame({'A': [1, 2, 3]}) after importing pandas as pd. Its input example uses a dictionary whose key supplies the column name and whose list supplies the values. This is a complete documentation example with its import included below. It gives a beginner a small named column before introducing file readers, joins or more complicated construction patterns.

import pandas as pd
pd.DataFrame({'A': [1, 2, 3]})

What we have not tested yet

Our November 22 run ended at example discovery, so it produced no completed analysis pipeline, application-correctness result or runtime benchmark. The documentation example above was not executed in that run. The November 11 source-install timeout has no raw failure tail supporting a narrower diagnosis, and no subsequent build or tests ran. Neither recorded container image label establishes the actual interpreter or operating-system version; this also corrects the earlier review's unsupported Python 3.12 runtime statement.

The 101.3 MB measurement covers site-packages after one installation, rather than peak working memory or a universal deployment footprint. We have no measurements of joins, engine-conversion overhead, Arrow-backed dtype behavior, Excel round trips, Spark distribution or openpyxl comparisons. The historical all-NaN issue does not identify a fixing release here, and the pandas 3.0 PyArrow decision is not a separate dependency audit of 3.1.1. There is no vulnerability assessment behind this review.

Jodhan's comparison and the TrustRadius account are attributed reports rather than our benchmarks or representative adoption statistics. GitHub's November 22 metadata records a same-day push and 1,795 open issues and pull requests; those facts do not measure support response times. Setup, documentation coverage, community support and production reliability therefore retain unassessed editorial scores. For a subsequent workload evaluation, preserve expected outputs for missing and invalid data and compare the specific expensive operation on the dataset your application uses.

Alternatives

ProjectWhat it isPick it when
Polars gh↗An analytical query engine with DataFrame and SQL interfaces.Choose it when expensive transformations map to native table operations and downstream tools accept the result.
NumPy gh↗Numerical arrays used beneath pandas' labeled structures.Choose it when the task and its consumers need numerical arrays rather than labeled table operations.
DuckDB gh↗An analytical SQL database that runs inside an application.Choose it when SQL over CSV or Parquet expresses the analysis more directly.

What people are saying

  1. [github-trending] pandas-dev/pandas

Sources

  1. pandas README
  2. pandas User Guide
  3. pandas installation guide
  4. Dave Jodhan: pandas vs Polars
  5. TrustRadius reviewer account
  6. Polars README
  7. DuckDB README

More data reviews