pip is the default Python package manager and conda is a system package manager. conda is closely associated with Python and in particular, Python for use in the field of data science.

The main advantages of using conda, is it can install Python and non-Python packages. The Jupyter project for example is an abbreviation of Julia, Python et R. A conda environment can be created with Python and the R kernel irkernel - conda-forge and therefore Python and R packages, allowing both programming languages to be used in JupyterLab. This is more difficult to achieve using pip. On Linux, the third programming language Julia conda-forge julia can also be installed using conda, although the developers as far as I know never managed to add Julia to conda for Windows. The general idea was to add a way to installed other dependencies used by data science in conda such as TeX miktex - conda-forge and codecs conda-forge ffmpeg which are used for example in some matplotlib plots and animations. There is a limited degree of success in such areas.

Some data science IDEs such as Spyder Spyder Release Notes recommend their standalone installer or a conda environment with specific dependencies in order to work properly. The developers state that there is support for pip but that:

please be aware that pip installations are for advanced users with good knowledge of all Spyder dependencies. Because of that, all installation problems you encounter are expected to be solved by you.

One area of confusion when it comes to using conda is that there are multiple channels:

• -c conda-forge (open source community channel)
• -c anaconda (tainted commercial channel)

-c conda-forge is the community channel which is open source and has the latest version of packages and the largest number of packages. In general -c conda-forge should be used for creating custom environments.

-c anaconda is a tainted commercial channel which has a more limited set of packages and general older "more stable" versions of packages which the Anaconda company use as part of their Anaconda Python distribution which they charge for commercial use.

There are three conda based installers. One is open source:

• Miniforge which has a bootstrap base Python environment which contains the conda package manager and uses the community channel -c conda-forge by default. The base Python environment should be used only to update the conda package manager and other -c conda-forge environments should be made for other projects.

Generally you would create an environment using only packages from -c conda-forge. There are some additional specialised channels that you can use such as -c bio-conda, a channel used to group life-science packages. The -c bio-conda channel is designed for compatibility with the -c conda-forge channel.

Two are tainted by Anacondas licensing agreements:

• Miniconda which has a bootstrap base Python environment which contains the conda package manager and uses the commercial channel -c anaconda by default. The base Python environment should be used only to update the conda package manager and other -c anaconda or -c conda-forge channel environments should be made for other projects. Do not mix channels per project as it results in an unstable environment.

• Anaconda which has a large number of data science packages installed in the base Python environment and generally should be used "as is". The base Python environment also contains the conda package manager and uses the commercial channel -c anaconda by default. In the base Python environment you should only look to update the conda package manager (and anaconda-navigator). Updating the conda package manager using the -c anaconda should in turn update the Anaconda distribution. Although, often it is more reliable to uninstall Anaconda and reinstall Anaconda using the latest standalone installer. The conda package manager can be used to create other -c anaconda or -c conda-forge environments for other projects. Do not mix channels per project as it results in an unstable environment.

Mixing channels generally results in packages from the -c conda-forge channel being downgraded to "more stable" packages from -c anaconda channel. Generally this results in an unstable environment.