When you use conda to install Python, by default a base environment is created and packages are installed in that environment. However, in general you should almost never undertake project work in the base environment, especially if your project involves installing any custom Python packages. Instead, you should create a new environment for each project, and install the necessary packages and dependencies in that environment. This will help prevent software conflicts, and ensure that tools installed for one project will not break another project.
Note
The instructions below provide only the most basic steps to set up and use an environment. Much more extensive documentation on managing environments is available in the conda documentation itself.
Use the terminal (MacOS/Linux) or an Anaconda Prompt (Windows) to create an environment:
conda create --name your_environment_name
Be sure to replace your_environment_name with a suitable
name for the environment to create.
When conda asks you to proceed, type “y” or just hit enter:
proceed ([y]/n)?
This creates the your_environment_name environment. By default,
this new environment uses the same version of Python that you are
currently using. If you want a specific version of Python you can
request it explicitly:
conda create --name your_environment_name python=3.7
You can also create an environment with one or more specific packages installed, by giving them as well:
conda create --name your_environment_name python=3.7 numpy pandas
Clearly, if you have a lot of packages to install, this can become a long command, and a bit unwieldy to use. Fortunately, you can instead just describe the environment you want to create in a YAML file instead of doing so on the command line. To do so, you would get or create a YAML file that looks something like this:
name: your_environment_name
channels:
- conda-forge
- defaults
dependencies:
- python=3.7
- pip
- numpy>=1.15.4
- pandas>=0.23.4
- scipy>=1.1
- scikit-learn>=0.20.1
- networkx
- pip:
- specialty_package
And then create the environment using the file.
conda env create -f environment.yml
You may notice that the specialty_package in the environment.yml file
is installed using pip instead of conda. This is
strongly discouraged
if the package is also available from conda, but may be necessary to
install certain packages that are available only on PyPI.
In addition to creating environments from a file on your local file system, you can also create environments using remote definitions available on anaconda.org. For example, an environment specifically set up for this transportation tutorial (‘tt’ for short) can be installed using the command:
conda env create camsys/tt
Note
If you installed the “Miniconda” version of the anaconda package, you may need to install or update the conda and anaconda-client packages before the remote environment installation will work:
conda install -n base -c defaults conda anaconda-client
The set of packages in any existing environment can be changed after creating an environment, using the conda install command. For example, to add the latest scikit-learn package to the tt environment,
conda install scikit-learn -n tt
You can also install multiple packages at one time, by just listing them as well.
conda install scikit-learn scikit-image -n tt
The -t flag names the environment into which to install the package(s). If this is omitted, the packages will be installed into the currently active environment. You can also install packages in the currently active environment using pip, although like in the original environment definition, it is recommended to install using conda if possible. It is also recommended to install as much as possible in the original definition and creation of the environment. Adding packages later usually works, but sometimes it can break functionality of previously installed packages (usually by changing some dependency). Conda does as best it can to prevent these problems, but inter-package dependencies are complicated to manage.
When using the terminal (MacOS/Linux) or an Anaconda Prompt (Windows), the current environment name will be shown as part of the prompt:
(base) C:\Users\cfinley>
(base) Computer:~ cfinley$
By default, when opening a new terminal the environment is set as the
base environment, although this is typically not where you want to
be if you have followed the advice above. Instead, to switch environments
use the conda activate command:
(base) C:\Users\cfinley> conda activate your_environment_name
(your_environment_name) C:\Users\cfinley>
(base) Computer:~ cfinley$ conda activate your_environment_name
(your_environment_name) Computer:~ cfinley$