Ctrl+K

Using Anaconda

Using Anaconda#

Installing Anaconda#

Last revised: 18-Dec-2023 by Dick Furnstahl (furnstahl.1@osu.edu)

Overview of what you need: Anaconda Python

You may already have Python on your computer, but it may not be the version we need (Python changes rapidly) and there are many additional software libraries we will need for scientific computing: Numpy, Scipy, Matplotlib are examples. Fortunately, there is a one-stop-shopping solution: download and install Anaconda Python. It comes with all of the packages we need, plus a great mechanism (a package manager called conda) for updating or installing new packages. Anaconda is available for Windows, macOS, and Linux. One caveat is that the package is not small: the size is over 3GB. If that is a problem, ask about other options to installing the full package (see the documentation on Miniconda).

How to get Anaconda Python [Also see the Anaconda Distribution Starter Guide]

Step-by-step instructions for downloading and installing Anaconda are available for Windows, Mac OSX, or Linux. Here we’ll step through the relevant options for us (based on following the graphical installer, which is what you will get by default for Windows and Macs; for Linux it is assumed you can follow the instructions for Linux).

  1. Go to https://www.anaconda.com/products/individual - Downloads and click on the Download button for your type of computer (use the 64-bit installer).

  2. An installation program will be downloaded. Start it up (e.g., by double clicking).

  3. Accept the license agreement.

  4. Install for All Users (this assumes you have admin privileges; it avoids potential later problems if you can do this). Allow the installation if your virus protection software asks you.

  5. It is suggested you accept the suggestion for the installation location (e.g., C:\ProgramData\Anaconda3 on Windows). [Note: steps 4 and 5 may be reversed for a Mac install.]

  6. For Windows, accept the checked Advanced Options as “Register Anaconda as my default Python x.x” and click Install.

  7. It is recommended (for our purposes) that you “Skip” installing Microsoft VSCode, if you are given the option, but this is your call.

  8. While it is installing, you might want to click Show Details to see progress. It will take a while, depending on your hardware. Be patient!

  9. If it works, you’ll get to a “Thanks for installing Anaconda Individual Edition!” page. There will be two already-checked boxes. You _ do not _ need to sign up for Anaconda Cloud, so uncheck that one. But you might want to read the “Getting started with Anaconda” information. Click Finish you are done.

Anaconda and github#

Installation of the LearningFromData notes and Jupyter notebooks from GitHub by command line

Go to the location where you want the notebook files.

Download the LearningFromData repository from GitHub and enter the directory LearningFromData which should have been created on your computer with the following:

git clone https://github.com/buqeye/LearningFromData.git
cd LearningFromData

The notebooks that we will be using depend on several scientific python modules (see the list in environment.yml) and require a python3.x installation based on Anaconda. This is best done within a conda environment, as described in the next section.

Anaconda environments#

Creating a conda environment#

The python modules needed for Learning from Data and their dependencies are best installed using conda by creating a virtual environment:

conda env create

which reads the environment.yml file in your current directory (in this case the one in LearningFromData). [Note: if you are using Windows, you should do this in an Anaconda Prompt window.]

To use a different file name such as other_environment.yml: conda env create -f other_environment.yml

It will take a while to generate the new environment; be patient! You will see a listing being generated like:

   Downloading and Extracting Packages
   scikit-learn-0.21.3  | 5.9 MB    | ##################################### | 100% 
   glib-2.58.3          | 3.1 MB    | ##################################### | 100% 
   libcxx-8.0.1         | 1000 KB   | ##################################### | 100% 
   scipy-1.3.1          | 18.1 MB   | ##################################### | 100%

etc. (it will be a pretty long list with different version numbers and not necessarily be in this order). Then you’ll see

   Preparing transaction: done
   Verifying transaction: done
   Executing transaction: done

where “done” appears when it has finished. If all is well at the end, you’ll get a success message. If it fails, email the message you get to furnstahl.1@osu.edu and we’ll figure out the problem.

Some packages might not be found in the default conda channels. One can either specify relevant package channel(s) in the environment.yml file (as done here), or add them to the default conda channel configuration via, e.g,

conda config --append channels conda-forge

You shouldn’t need to do this for the initial setup.

Once the virtual environment has been created it can be activated (the name “8820-env” was specified in the environment.yml file):

conda activate 8820-env

To deactivate the virtual environment:

conda deactivate

Note that earlier versions of conda used ‘source’ instead of ‘conda’ to activate environments. If you have an earlier version of conda you might have to do

conda update conda

Note that there are also other options (‘venv’, ‘pipenv’) for creating virtual environments that includes the python version and packages that we will be using.

Once the environment is set up and activated, you are encouraged to run the test Jupyter notebook:

cd LectureNotes/notebooks
jupyter notebook Anaconda_Bayes_test.ipynb

Other notebooks can be found in the subdirectories of the notebooks directory.

Updating your conda environment for Learning from Data#

Go to the LearningFromData directory you created by cloning the class repository from GitHub. This is where the relevant environment.yml file is stored. This file defines the environment and will be occasionally updated to add additional modules.
You can update to the new environment with:

conda deactivate
conda env update

Now if you activate the environment again:

conda activate 8820-env

you will have access to the new packages.

Changing to the 8820-env env kernel when running a Jupyter notebook#

If you are running the 8820-env kernel, you should see

Python [conda env:8820-env] *

in the upper right-hand corner of the Jupyter notebook you are running. If it just says something like Python 3 then you are not running the 8820-env kernel. In that case, look under the Kernel menu to Change kernel and select Python [conda env:8820-env]. The kernel should restart and indicate the new kernel in the upper corner.

Contents