# Setting up an Environment¶

## Anaconda¶

Creating environments using Anaconda is recommended due to the ability to create more than one environment. It is also recommended because you can keep dependencies separate from one another that might conflict if you had them all in your root environment. For example, if you had all the dependencies for a Pandas environment and all the dependencies for a Py-ART environment in your root environment, there might be conflicts between channels and packages. So Anaconda allows you to create multiple environments to avoid these issues.

While Anaconda is downloading, it will ask if you want to set a path to it, or let Anaconda set a default path. After choosing, Anaconda should finish downloading. After it is done, exit the terminal and open a new one to make sure the environment path is set. If conda command is not found, there is help on running conda and fixing the environment path, found here:

## Setting a Channel¶

Anaconda has a cloud that stores many of its packages. It is recommended, at times, to use the conda-forge channel instead. Conda-Forge is a community led collection of packages, and typically contains the most recent versions of the packages required for Py-ART. Also Py-ART is on Conda-Forge. Having packages in an environment, within the same channel, helps avoid conflict issues. To add conda-forge as the priority channel, simply do:

conda config --add channels conda-forge


You can also just flag the channel when conda install packages such as:

conda install -c conda-forge numpy


More on managing channels can be found here:

## Creating an Environment¶

There are a few ways to create a conda environment for using Py-ART or other packages. One way is to use the environment file, found here:

To create an environment using this file, use the command:

conda env create -f environment.yml


This will then create an environment called pyart_env that can be activated by:

source activate pyart_env


or deactivated after use:

source deactivate pyart_env


Once the environment is created and activated, you can install more packages into the environment by simply conda installing them. An example of this is, if you want Jupyter Notebook to run in that enviroment with those packages:

conda install -c conda-forge jupyter notebook


while that environment is activated. Another way to create a conda environment is by doing it from scratch using the conda create command. An example of this:

conda create -n pyart_env -c conda-forge python=3.6 arm_pyart netCDF4
cartopy scipy numpy matplotlib


This will also create an environment called pyart_env that can be activate the same way, as mentioned above. To then run your coding editor within the environment, run in the command line:

python


or:

ipython


or:

jupyter notebook


or even:

spyder


depending on what you installed in your environment and want to use for coding.

## Adding Optional Dependencies with setting Paths¶

There are other optional dependencies that can enhance the use of Py-ART. One, such package is CyLP. To get CyLP to work, installing of the package coincbc is needed as a dependency for CyLP. Simply do:

conda install -c conda-forge coincbc


within your pyart_env. After that though, the coincbc path needs to be exported so CyLP knows where to find it during its install. To do this:

export COIN_INSTALL_DIR=/Users/yourusername/youranacondadir/envs/pyart_env


or real example on a Linux machine:

export COIN_INSTALL_DIR=/home/zsherman/anaconda3/envs/pyart_env


CyLP was actually adapted by Jonathan Helmus to be Python 3 compatible, so we will install a specific CyLP branch after doing the export path step above. GitHub repositories can actually be pip installed within your environment. So to install the CyLP version we want:

pip install git+https://github.com/jjhelmus/CyLP.git@py3


This will install a Python 3 compatible version of CyLP found on GitHub.