Skip to main content

​
Requirements

  • Python 3.8+ (64-bit)
  • pip 20.3+
datatable has no dependencies on other Python packages or non-standard system libraries.

​
Install with pip

1

Upgrade pip

Make sure you are running pip 20.3 or later. Older versions will cause a BackendUnavailable error.
pip install pip --upgrade
2

Install datatable

Install the latest stable release from PyPI:
pip install datatable
Binary wheels are provided for the following platforms:
PlatformDetails
macOSTested on macOS 10.12 (Sierra) through macOS 12 (Monterey)
Linux x86_64 / ppc64lemanylinux_2_17-compatible distributions (see PEP 600)
WindowsWindows 10 or later
3

Verify the installation

Confirm datatable installed correctly:
import datatable as dt
print(dt.__version__)
You should see the installed version number printed.

​
Install a development version

If you want to test a pre-release build before it is officially published:
Development wheels are published to an S3 repository. Browse https://h2o-release.s3.amazonaws.com/datatable/index.html, find the wheel for your Python version and platform, then install it:
pip install YOUR_WHEEL_URL

​
Build from source

Use this path if you are on a platform without a prebuilt wheel, or if you want to contribute to datatable.

​
Install the latest source from GitHub

pip install git+https://github.com/h2oai/datatable
datatable is written mostly in C++, so your system needs a working C++ compiler. The build script searches for GCC, Clang, or MSVC automatically. To use a different compiler, set the CXX environment variable before running the install. The minimum supported compiler versions are:
  • Clang 5+
  • GCC 6+
  • MSVC 19.14+
datatable uses the C++14 language standard.

​
Set up an editable development install

If you want to modify datatable or add your own functionality:
1

Clone the repository

Fork the repo on GitHub, then clone your fork:
git clone https://github.com/your_user_name/datatable
cd datatable
2

Build the core library

# Production build
make build

# Debug build (no optimizations, internal asserts enabled)
make debug
On macOS you may need Xcode Command Line Tools: run xcode-select --install. On Linux, if you see 'Python.h' file not found, install the development headers for your Python version, e.g. python3.8-dev.
A successful build produces a _datatable.*.so file in src/datatable/lib/.
3

Register the source tree with Python

echo "`pwd`/src" >> ${VIRTUAL_ENV}/lib/python*/site-packages/easy-install.pth
Adjust the path if you are not using a virtualenv.
4

Install test and extra dependencies

pip install -r requirements_tests.txt
pip install -r requirements_extra.txt
pip install -r requirements_docs.txt
5

Run the test suite

make test
After the initial setup, subsequent development is simpler: re-run make build (or make debug) after any C++ change, then restart Python. Only modified files are recompiled, so incremental builds usually take a few seconds.

​
Troubleshooting

pip._vendor.pep517.wrappers.BackendUnavailable Your pip version is too old. Upgrade to 20.3+ with pip install pip --upgrade. ImportError: cannot import name '_datatable' The internal _datatable.*.so core library is missing, in the wrong location, or has the wrong name. Reinstall datatable. If the file is present but not under site-packages/datatable/lib/, move it there. If the suffix doesn’t match what Python expects, check it with: 
import sysconfig
sysconfig.get_config_var("SOABI")
Python.h: no such file or directory (source builds on Linux) Install the Python development package: e.g., sudo apt install python3.8-dev. fatal error: 'sys/mman.h' file not found (source builds on macOS) Install Xcode Command Line Tools: xcode-select --install. ImportError: This package should not be accessible A misconfigured PYTHONPATH environment variable is shadowing the installation. Unset it and try again.
For questions not covered here, ask on Stack Overflow using the [py-datatable] tag, or file an issue on GitHub.

Build docs developers (and LLMs) love

Get started for freeTalk to us