Virtual environment

Sometimes we would like to write code with a specific Python version and/or library because, for example, some libraries work only on Python 3.5, not 3.6. Then we need to create a virtual environment for the project.

The following lines are supposed to be run in Anaconda prompt on Windows or a shell on Mac.

Creating a virtual environment

conda create -n NAME_OF_ENVIRONMENT python=3.5 anaconda

We can choose the name of environment arbitrary. Then, on Windows, a corresponding Anaconda prompt like "Anaconda prompt (***)" would be created. The environment can be used easily on Jupyter Notebook and PyCharm.

Deleting a virtual environment

conda remove -n NAME_OF_ENVIRONMENT --all

Checking environments

conda info -e

Activate an environment

activate NAME_OF_ENVIRONMENT

We are supposed to use "source activate ***" on Mac OS, but this line produces an error like "pyenv: -bash: command not found" possibly due to some conflict between pyenv and conda (2017.04.11). Then we need to specify the full path as

source ~/.pyenv/versions/anaconda3-2.5.0/bin/activate ***

Note that we should modify the line according to the position of pyenv root folder and the version of Anaconda.

Deactivate an environment

deactivate NAME_OF_ENVIRONMENT

We should use "source deactivate ***" on Mac OS.

Cloning current environment

conda create -n *** --clone root

By doing this, we can make a backup (***) before updating the current environment (root).

Using an environment in PyCharm

• Open your project of interest in PyCharm
• Open Preferences/Settings -> project interpreter tab
• Click the cog icon and select "Add local"
• Select the desired Python executable located at, for example, ~/.pyenv/versions/anaconda3-2.5.0/envs/***/bin/python

Using an environment with Jupyter Notebook

• Open Jupyter Notebook in the virtual environment

(2017.7.7) I found that, when using jupyter notebook in multiple virtual environments, the kernel specification would be broken. Although I am not sure how this happens, the following procedure fixed this problem.

• Open a jupyter notebook in an arbitrary environment
• Locate the kernel spec file by "from jupyter_core.paths import jupyter_data_dir; print(jupyter_data_dir())"
• Open "kernel.json"
• replace the absolute path to an executable "/Users/***/python" by "python"