The Intel® Distribution for Python* provides accelerated performance to some of the most popular packages in the Python ecosystem, and now select packages have the added the option of installing from the Python Package Index (PyPI) using pip. The packages require the use of pip version 9.0.1, and are available utilizing the following instructions:
Performance Packages
The two most popular packages in numerical and scientific work (numpy and scipy) are available with the following commands below. For more information on the nature of their accelerations and performance benchmarks, please visit the link here.
Also, Intel-optimized-scikit-learn, pydaal(Intel® DAAL in Python) and tbb4py(Intel® TBB for Python) are also available on PyPI now.
| Package Name | pip command | Platform Availability |
| numpy | pip install intel-numpy | Linux, Win, macOS(10.12) |
| scipy | pip install intel-scipy |
| scikit-learn | pip install intel-scikit-learn |
| pydaal | pip install pydaal |
| tbb4py | pip install tbb4py |
Optimized Python packages such as intel-scikit-learn, intel-scipy and pydaal utilize intel-numpy.
Based on PyPI's dependency resolution on Intel variants, If one installs intel-numpy, one would also get mkl_fft and mkl_random (with NumPy). Similarly, if one installs intel-scipy, one would also get intel-numpy along with SciPy. And, if one installs intel-scikit-learn, one would also get intel-numpy,intel-scipy along with Scikit-Learn.
Note: If standard NumPy, SciPy and Scikit-Learn packages are already installed, the packages must be uninstalled before installing the Intel® variants of these packages(intel-numpy etc) to avoid any conflicts. As mentioned earlier, pydaal uses intel-numpy, hence it is important to first remove the standard Numpy library(if installed) and then install pydaal.
To uninstall existing packages, run the command:
pip uninstall numpy scipy scikit-learn -y
Specialized NumPy packages
Several specialized Intel packages act as a complement to numpy and scipy, which provide accelerated Fast Fourier Transforms and improved Random functionality through the MKL when paired with numpy and scipy.
| Package Name | pip command | Platform Availability |
| mkl_fft | pip install mkl_fft | Linux, Win, macOS(10.12) |
| mkl_random | pip install mkl_random |
Note: In order to utilize these packages, the standard NumPy installation must be removed first using the command: pip uninstall numpy -y
Intel® Runtime Packages
The runtime packages are built runtime distributable libraries that allow for dispatch of vectorization on Intel hardware. For Python packages that depend on these runtimes, they can be individually downloaded as well. For more information, please visit the link here.
| Package Name | pip command | Platform Availability |
| mkl | pip install mkl | Linux, Win, macOS(10.12) |
| ipp | pip install ipp |
| daal | pip install daal |
| intel-openmp | pip install intel-openmp |
| tbb | pip install tbb |
| impi | pip install impi | Linux, Win |
Development only packages
For those building their own Python packages with Intel® Parallel Studio XE or building and linking with the Intel® Performance Libraries, the devel packages assist in providing the development runtimes pre-built for testing, and are available with the following commands:
| Package Name | pip command | Platform Availability |
| mkl-devel | pip install mkl-devel | Linux, Win, macOS(10.12) |
| ipp-devel | pip install ipp-devel |
| daal-devel | pip install daal-devel |
Troubleshooting
While `pip install`-ing any package, if installation fails with the following error message :
zlib.error: Error -5 while decompressing data: incomplete or truncated stream
retry after running the following command: rm -rf ~/.cache/pip
