Installation

sktime currently supports:

  • environments with python version 3.6, 3.7, or 3.8.

  • operating systems Mac OS X, Unix-like OS, Windows 8.1 and higher

See here for a full list of precompiled wheels available on PyPI.

We appreciate community contributions towards compatibility with python 3.9, or other operating systems.

Release versions

For frequent issues with installation, consult the Release versions - troubleshooting section.

Installing sktime from PyPI

sktime releases are available via PyPI and can be installed via pip using:

pip install sktime

This will install sktime with core dependencies, excluding soft dependencies.

To install sktime with maximum dependencies, including soft dependencies, install with the all_extras modifier:

pip install sktime[all_extras]

Installing sktime from conda

sktime releases are available via conda from conda-forge. They can be installed via conda using:

conda install -c conda-forge sktime

This will install sktime with core dependencies, excluding soft dependencies.

To install sktime with maximum dependencies, including soft dependencies, install with the all-extras recipe:

conda install -c conda-forge sktime-all-extras

Note: currently this does not include dependencies catch-22, pmdarima, and tbats. As these packages are not available on conda-forge, they must be installed via pip if desired. Contributions to remedy this situation are appreciated.

Release versions - troubleshooting

Module not found

The most frequent reason for “module not found” errors is installing sktime with minimum dependencies and using an estimator which interfaces a packages that has not been installed in the environment. To resolve this, install the missing package, or install sktime with maximum dependencies (see above).

Facebook prophet

A frequent issue arises with installation of facebook prophet when installing via pip, especially on Windows systems.

Potential workaround no.1, install manually via conda-forge:

conda install -c conda-forge pystan
conda install -c conda-forge prophet

The remaining packages can be installed via pip or conda.

Potential workaround no.2, install pystan with no-cache parameter:

pip install pystan --no-cache

Potential workaround no.3, if on Windows: use WSL (Windows Subsystem for Linux), see end of section Windows 8.1 and higher.

Development versions

To install the the latest development version of sktime, or earlier versions, the sequence of steps is as follows:

Step 1 - git clone the sktime repository, the latest version or an earlier version. Step 2 - ensure build requirements are satisfied Step 3 - pip install the package from a git clone, with the editable parameter.

Detail instructions for all steps are given below. For brevity, we discuss steps 1 and 3 first; step 2 is discussed at the end, as it will depend on the operating system.

Step 1 - git cloning the repository

The sktime repository should be cloned to a local directory, using a graphical user interface, or the command line.

Using the git command line, the sequence of commands to install the latest version is as follows:

  1. Clone the repository: git clone https://github.com/alan-turing-institute/sktime.git

  2. Move into the root directory of the local clone: cd sktime

  3. Make sure you are on the main branch: git checkout main

  4. Make sure your local version is up-to-date: git pull

To build a previous version, replace line 3 with:

git checkout <VERSION>

This will checkout the code for the version <VERSION>, where <VERSION> is a valid version string. Valid version strings are the repository’s git tags, which can be inspected by running git tag.

You can also download a zip archive of the version from GitHub.

Step 2 - satisfying build requirements

Before carrying out step 3, the sktime build requirements need to be satisfied. Details for this differ by operating system, and can be found in the sktime build requirements section below.

Typically, the set-up steps needs to be carried out only once per system. That is, the steps usually do not need to be followed again on the same system when installing an sktime development version for the second or third time. Similarly, the advanced developer set-up requires the same build requirements, so typically no additional steps are required if the advanced developer set-up has already been completed.

Step 3 - building sktime from source

For a static install of sktime from source, navigate to the local clone’s root directory and type:

pip install .

Alternatively, the . may be replaced with a full or relative path to the root directory.

For a developer install that updates the package each time the local source code is changed, install sktime in editable mode, via:

pip install --editable .

This allows editing and extending the code in-place. See also the pip reference on editable installs).

Note

You will have to re-run:

pip install --editable .

every time the source code of a compiled extension is changed (for instance when switching branches or pulling changes from upstream). Compiled extensions are Cython files (ending in .pyx or .pxd).

Building binary packages and installers

The .whl package and .exe installers can be built with:

pip install wheel
python setup.py bdist_wheel

The resulting packages are generated in the dist/ folder.

sktime build requirements

This section outlines the sktime build requirements. These are required for:

  • installing sktime from source, e.g., development versions

  • the advanced developer set-up

Build requirements summary

The core build requirement for sktime are:

  • Cython >= 0.28.5 (available through pip install cython)

  • OpenMP and a working C compiler (see below for instructions)

Note

It is possible to build sktime without OpenMP support by setting the SKTIME_NO_OPENMP environment variable (before cythonization). This is not recommended since it will force some estimators to run in sequential mode and their n_jobs parameter will be ignored.

For the advanced developer set-up which includes tests and documentation, see the advanced developer documentation in How to contribute.

The following sections describe how to satisfy the build requirements, by operating system. We currently support:

FreeBSD

The clang compiler included in FreeBSD 12.0 and 11.2 base systems does not include OpenMP support. You need to install the openmp library from packages (or ports):

sudo pkg install openmp

This will install header files in /usr/local/include and libs in /usr/local/lib. Since these directories are not searched by default, you can set the environment variables to these locations:

export CFLAGS="$CFLAGS -I/usr/local/include"
export CXXFLAGS="$CXXFLAGS -I/usr/local/include"
export LDFLAGS="$LDFLAGS -L/usr/local/lib -lomp"
export DYLD_LIBRARY_PATH=/usr/local/lib

Finally you can build the package using the standard command.

For the upcoming FreeBSD 12.1 and 11.3 versions, OpenMP will be included in the base system and these steps will not be necessary.

Mac OSX

The default C compiler, Apple-clang, on Mac OSX does not directly support OpenMP. The first solution to build sktime is to install another C compiler such as gcc or llvm-clang. Another solution is to enable OpenMP support on the default Apple-clang. In the following we present how to configure this second option.

You first need to install the OpenMP library:

brew install libomp

Then you need to set the following environment variables:

export CC=/usr/bin/clang
export CXX=/usr/bin/clang++
export CPPFLAGS="$CPPFLAGS -Xpreprocessor -fopenmp"
export CFLAGS="$CFLAGS -I/usr/local/opt/libomp/include"
export CXXFLAGS="$CXXFLAGS -I/usr/local/opt/libomp/include"
export LDFLAGS="$LDFLAGS -L/usr/local/opt/libomp/lib -lomp"
export DYLD_LIBRARY_PATH=/usr/local/opt/libomp/lib

Finally you can build the package using the standard command.

Troubleshooting - Mac OSX build requirements

After installing the release version following the installation steps above and running a pytest command, some contributors received the error message below:

E   ValueError: numpy.ndarray size changed, may indicate binary incompatibility. Expected 88 from C header, got 80 from PyObject

A possible solution to the problem is reinstalling your C compiler. If it is gcc, run brew reinstall gcc and then pip install -e .. This should be followed by installing the OpenMP library and setting the environment variables using the same commands again as in the section above.

If you found another solution to the problem, please kindly consider contributing to this section.

Unix-like OS

Installing from source without conda requires you to have installed the sktime runtime dependencies, Python development headers and a working C/C++ compiler. Under Debian-based operating systems, which include Ubuntu:

sudo apt-get install build-essential python3-dev python3-setuptools \
                 python3-pip

and then:

pip3 install numpy scipy cython

When precompiled wheels are not available for your architecture, you can install the system versions:

sudo apt-get install cython3 python3-numpy python3-scipy python3-matplotlib

On Red Hat and clones (e.g. CentOS), install the dependencies using:

sudo yum -y install gcc gcc-c++ python-devel numpy scipy

Note

To use a high performance BLAS library (e.g. OpenBlas) see scipy installation instructions.

Windows 8.1 and higher

To build sktime on Windows you need a working C/C++ compiler in addition to numpy, scipy and setuptools.

The building command depends on the architecture of the Python interpreter, 32-bit or 64-bit. You can check the architecture by running the following in cmd or powershell console:

python -c "import struct; print(struct.calcsize('P') * 8)"

The above commands assume that you have the Python installation folder in your PATH environment variable.

You will need Build Tools for Visual Studio 2017.

Warning

You DO NOT need to install Visual Studio 2019. You only need the “Build Tools for Visual Studio 2019”, under “All downloads” -> “Tools for Visual Studio 2019”.

For 64-bit Python, configure the build environment with:

SET DISTUTILS_USE_SDK=1
"C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\VC\Auxiliary\Build\vcvarsall.bat" x64

Please be aware that the path above might be different from user to user. The aim is to point to the “vcvarsall.bat” file.

And build sktime from this environment:

python setup.py install

Replace x64 by x86 to build for 32-bit Python.

Some users have experienced issues when installing NumPy, particularly version 1.19.4. Note that a recent Windows update may affect compilation using Visual Studio (see Windows update issue).

If you run into a problem installing the development version and are using Anaconda, try:

  1. Install Anaconda

  2. Create new environment: conda create -n sktime-dev python=3.8

  3. Activate environment: conda activate sktime-dev

  4. Install NumPy (pinned to 1.19.3) from pip: pip install numpy==1.19.3

  5. Install requirements: pip install -r build_tools/requirements.txt

  6. Follow the instructions above to point to “vcvarsall.bat”

  7. Run pip install --verbose --no-build-isolation --editable .

In step 5, you may optionally install the packages in build_tools/requirements.txt that are available from Anaconda’s default channels or Conda-Forge via Conda. Any remaining packages can be added via pip.

Note

It is possible to use MinGW (a port of GCC to Windows OS) as an alternative to MSVC for 32-bit Python. Not that extensions built with mingw32 can be redistributed as reusable packages as they depend on GCC runtime libraries typically not installed on end-users environment.

To force the use of a particular compiler, pass the --compiler flag to the build step:

python setup.py build --compiler=my_compiler install

where my_compiler should be one of mingw32 or msvc.

Note

Another option under Windows is to follow the instructions for Unix-like OS, using the Windows Subsystem for Linux (WSL). For installing WSL, follow the instructions here.

References

The installation instruction are adapted from scikit-learn’s advanced installation instructions.