Conda, Mamba & Pip#

Tip

Use mamba (repo, doc) for faster dependency resolution during package installation

conda Command Reference#

  • Install mamba into the conda base environment:

    $ conda install mamba -n base -c conda-forge

  • To update conda itself, do something like this:

    $ conda update -n base -c defaults conda

Environment Activation, Inspection & Deactivation#

  • Show environments available:

    $ conda env list

  • Activate an environment:

    $ conda activate <env_name>

  • Deactivate the current environment:

    $ conda deactivate

  • List all packages (and their versions) in the current environment

    • Option 1 – Easily readable by eye

      $ conda list

      creates output that looks like this:

      # packages in environment at /Users/drauch/Software/Conda/miniconda/miniconda3/envs/spellbook:
#
# Name                    Version                   Build  Channel
abseil-cpp                20210324.2           h23ab428_0
absl-py                   0.15.0             pyhd3eb1b0_0
aiohttp                   3.8.1            py39hca72f7f_0
aiosignal                 1.2.0              pyhd3eb1b0_0
alabaster                 0.7.12             pyhd3eb1b0_0

# [...] output truncated

    • Option 2 – For use by packaging tools

      $ conda list -e

      creates the following output:

      # This file may be used to create an environment using:
# $ conda create --name <env> --file <this file>
# platform: osx-64
abseil-cpp=20210324.2=h23ab428_0
absl-py=0.15.0=pyhd3eb1b0_0
aiohttp=3.8.1=py39hca72f7f_0
aiosignal=1.2.0=pyhd3eb1b0_0
alabaster=0.7.12=pyhd3eb1b0_0

# [...] output truncated

      This can be used to create a file that can later be used to recreate the same environment:

      $ conda list -e > conda-requirements.txt
$ conda create --name <env> --file conda-requirements.txt

      Note that the conda-requirements.txt file does not work with pip, though. For pip, you should do

      $ pip freeze > requirements.txt
$ pip install -r requirements.txt

Environment Creation & Update#

  • Create an environment and install packages:

    $ conda create --name <env_name> python=3.8 tensorflow

  • Create environment from requirements file:

    $ conda env create --file spellbook.yml

    or - if mamba is installed -

    $ mamba env create --file spellbook.yml

  • Update an environment after changes to the requirements file:

    $ conda env update --file spellbook.yml --prune

    or - if mamba is installed -

    $ mamba env update --file spellbook.yml --prune

  • Register an environment with Jupyter notebooks:

    $ python -m ipykernel install --user --name=<env_name>

  • List the environments registered with Jupyter:

    $ jupyter kernelspec list

Environment Deletion#

  • Delete an environment:

    $ conda env remove -n <env_name>

  • Remove an environment from Jupyter notebooks:

    $ jupyter kernelspec uninstall <env_name>

pip Command Reference#

  • Install a package from a directory in editable mode, e.g. for development. This requires a setup.py:

    $ pip install -e .

    or

    $ pip install -e some/folder

Mixing conda and pip#

General guidelines:

  • First use conda, then pip

  • When a conda environment is activate, pip will install packages into it

  • If something goes wrong with pip, better don’t bother trying to fix it, just delete the entire environment and reinstall from the conda yaml file and the pip requirements.txt

You can add pip packages to be installed into the conda yaml file - when the environment is created, pip is run and the specified packages are installed:

name: my-env-1
channels:
    - conda-forge
    - anaconda

dependencies:
    - python = 3.7
    - ipykernel # required to register the env as a kernel
    - numpy = 1.20.2

    - pip
    - pip:

       # add an additional package index / package repository
       # - --extra-index-url https://some-server.url/with/a/package/repository

       # add some packages
       - package1
       - package2