Tip: Using Conda Environments with Jupyter on Windows

Introduction

Conda environments don’t work well by default in PowerShell, and the environments don’t show up by default as kernels in Jupyter.

Using Conda Environments in PowerShell

Update 1/29/2019: Conda now officially supports PowerShell. Read more on the blog post.

The ease with which Conda environments can be used in PowerShell varies based on the version of Conda installed. Check your version by running conda --version.

Version 4.6 and Later

With the version 4.6 update, Conda supports PowerShell natively. To configure for PowerShell, run the following:

PS> conda init powershell

The conda init command will modify the activate/deactivate scripts and make changes to your PowerShell profile.

Prior to Version 4.6

You’ll need to install some additional software in order to use the activate/deactivate commands in PowerShell. Specifically, install the helpful tool from Pavel Koneski called PSCondaEnvs.

Note that if your PowerShell execution policy is Restricted (run Get-ExecutionPolicy to check), you’ll need to set it to RemoteSigned by running the following:

PS> Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned

Now, install pscondaenvs with Conda:

PS> conda install -n root -c pscondaenvs pscondaenvs

Enter y when prompted to proceed.

Once pscondaenvs is installed, you’ll have new activate and deactivate commands available that work in PowerShell.

Making Conda Environments Available in Jupyter

First, activate the desired environment by running conda activate myenv (or activate myenv if using pscondaenvs). Ensure dependencies are available by running the following:

(myenv) PS> conda install pip
(myenv) PS> conda install ipykernel

Type y to proceed as needed. After the installation completes, run the following command to create a new ipykernel for this environment:

(myenv) PS> python -m ipykernel install --user --name myenv --display-name "Python (myenv)"

You’ll see the “Python (myenv)” kernel available when you next launch JupyterLab.