Scikit-learn is a free software machine learning library for the Python programming language. It features various classification, regression and clustering algorithms including support vector machines, random forests, gradient boosting, k-means and DBSCAN, and is designed to interoperate with the Python numerical and scientific libraries NumPy and SciPy.
The installation process on Linux Mint and other Debian derivated systems is very simple, but there is a final step that is ambiguous which is the re-configuration of the Python interpreter to load the provided Anaconda libraries which includes the "scikit-learn" package.
Installing Anaconda is as easy as going to the download page, choosing the appropriate version for your python installation, following the instructions step-by-step, and letting the installer do its thing. In my case, in order to keep things simple, I decided to change the target directory to
/opt/anaconda/ and at the end of the installation process the script suggests to include the new binary path into the global PATH environment variable.
After this you have to restart your current Terminal or execute
source ~/.bashrc to load the new configuration. At this point we expect to have all the Anaconda-related programs loaded system wide, we can check this executing
And here is where it gets tricky...
From the Scikit-learn website we can read that the latest version of Canopy and Anaconda both ship a recent version of scikit-learn, in addition to a large set of scientific python library. We can verify this by requesting an update of the repository registry and trying to install the "sciki-learn" package, I say "try" because it is supposed to be already installed.
$ ls /opt/anaconda/pkgs/ $ conda update conda $ conda install scikit-learn
However, trying to load the "sklearn" module fails:
$ python -c "import sklearn" Traceback (most recent call last): File "<string>", line 1, in <module> ImportError: No module named 'sklearn'
This is why I decided to write this article, maybe someone will get stuck in this step the same way I was. Here is what happened... Before you installed Anaconda you were provided two options for the installer, one for Python3.x and one for Python2.x, this is important because after the installation a new set of aliases are installed in
/opt/anaconda/bin/python* which are supposed to replace the interpreter shipped by default with your operating system.
If we execute the aliases inside the Anaconda directory the module will load:
$ /opt/anaconda/bin/python -c "import sklearn" $ if [[ "$?" -eq 0 ]]; then echo "Success"; else echo "Failure"; fi
So in order to make this alias the default entry point for all our future Machine Learning projects we will have to force the system to load the interpreter inside the Anaconda directory system wide. A simple re-linking of
/usr/bin/python should work but fortunately Linux Mint (and other Debian derived systems) come with a tool named
$ sudo -s # Root privileges are required $ update-alternatives --install /usr/bin/python python /opt/anaconda/bin/python 1 $ python --version Python 2.7.12 :: Anaconda 4.1.1 (64-bit)
Now, we are all set... At least with the scikit-learn part. With the modification of the default binary that acts as the Python interpreter we just broke all the rest of the system LOL ¯_(ツ)_/¯ but there is an easy fix for that, we will load the default Python module directories into
export PATH="$PATH:/opt/anaconda/bin" export PYTHONPATH="$PYTHONPATH:/usr/lib/pymodules/python2.7" export PYTHONPATH="$PYTHONPATH:/usr/share/pyshared" export PYTHONPATH="$PYTHONPATH:/opt/anaconda/pkgs" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/plat-x86_64-linux-gnu" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/lib-tk" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/lib-old" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/lib-dynload" export PYTHONPATH="$PYTHONPATH:/usr/local/lib/python2.7/dist-packages" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/dist-packages" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/dist-packages/PILcompat" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/dist-packages/gst-0.10" export PYTHONPATH="$PYTHONPATH:/usr/lib/python2.7/dist-packages/gtk-2.0"
Check if the directories above are actually being loaded:
$ python -c "import sys, pprint; pprint.pprint(sys.path)"
And now we are all set, you can start hitting your head against the wall while learning about Machine Learning with Scikit-learn here are some videos that act as an introductory tutorial into this amazing world of data science:
- Hello World - Machine Learning Recipes #1
- Visualizing a Decision Tree - Machine Learning Recipes #2
- What Makes a Good Feature? - Machine Learning Recipes #3
- Let’s Write a Pipeline - Machine Learning Recipes #4
- Writing Our First Classifier - Machine Learning Recipes #5
- Train an Image Classifier with TensorFlow for Poets - Machine Learning Recipes #6
- Classifying Handwritten Digits with TF.Learn - Machine Learning Recipes #7