Try on Binder without local installation
Install on local machine with Conda
xESMF requires Python>=3.8. The major dependencies are xarray and ESMPy, and the best way to install them is using Conda. Note that the latest xarray releases require Python 3.9 or later.
First, install miniconda. Then, we recommend creating a new, clean environment:
$ conda create -n xesmf_env $ conda activate xesmf_env
Getting xESMF is as simple as:
$ conda install -c conda-forge xesmf
We also highly recommend those extra packages for full functionality:
# to support all features in xESMF $ conda install -c conda-forge dask netCDF4 # optional dependencies for executing all notebook examples $ conda install -c conda-forge matplotlib cartopy jupyterlab
Alternatively, you can first install dependencies, and then use
pip to install xESMF:
$ conda install -c conda-forge esmpy xarray numpy shapely cf_xarray sparse numba $ pip install git+https://github.com/pangeo-data/xesmf.git
This will install the latest version from the github repo. To install a specific release, append the version tag to the url (e.g. @v0.5.0).
Notes about ESMpy
ESMpy 8.4 is only compatible with xESMF >= 0.7.
ESMpy must be installed through Conda or compiled manually; it is not available through PyPI. When installing xESMF with pip, the ESMpy package must be manually installed first.
When installing ESMpy 8.4 through conda, the environment must be _activated_ before the package is used, which often means deactivating and re-activating it. See this xESMF issue and the related ESMpy issue.
Testing your installation
xESMF itself is a lightweight package, but its dependency ESMPy is a quite heavy and sometimes might be installed incorrectly. To validate & debug your installation, you can use pytest to run the test suites:
$ pip install pytest $ pytest -v --pyargs xesmf # should all pass
A common cause of error (especially for HPC cluster users) is that pre-installed modules like NetCDF, MPI, and ESMF are incompatible with the conda-installed equivalents. Make sure you have a clean environment when running
conda install (do not
module load other libraries). See this issue for more discussions.
Notes for Windows users
Installing scientific software on Windows can often be a pain, and Docker is a pretty good workaround. It takes some time to learn but worths the effort. Check out this tutorial on using Docker with Anaconda.
This problem is being investigated. See this other issue.
Install development version from GitHub repo
To get the latest version that is not uploaded to PyPI yet:
$ pip install --upgrade git+https://github.com/pangeo-data/xESMF.git
Developers can track source code change:
$ git clone https://github.com/pangeo-data/xESMF.git $ cd xESMF $ pip install -e .