Installation#

Earth-2 MIP is a Python package and can be installed from source using pip. However, we suggest setting up a dedicated run environment for Earth-2 MIP.

Installing from PyPi#

Presently Earth-2 MIP is not on PyPi. Install from source in the mean time.

Install from Release Wheel (Coming soon)#

Built Python wheels can be access in Earth-2 MIPs releases on Github. Navigate to the latest release of Earth-2 MIP and download the latest .whl. Install using pip:

pip install wheel_file.whl

Install from Source#

To install Earth-2 MIP from source:

git clone git@github.com:NVIDIA/earth2mip.git

cd earth2mip && pip install .

Earth-2 MIP Environments#

When running Earth-2 MIP, it is highly encourage to do so in a dedicated environment to mitigate the chance of unforseen package conflicts. Docker is the prefered environment, but if you are not familiar with docker or cannnot run a docker container Conda is typically a suitable alternative.

Conda Environment#

Start with installing Miniconda / Anaconda. Once installed we will create a new conda enviorment for Earth-2 MIP:

conda create --name earth2mip python=3.10
conda activate earth2mip

Next we can install Earth-2 MIP using one of the listed methods above. E.g.

git clone git@github.com:NVIDIA/earth2mip.git

cd earth2mip && pip install .

If you are interested in running Jupyter Notebooks (such as the provided examples), create a iPython kernel using the environment to use in Jupyter lab:

pip install ipykernel
python -m ipykernel install --user --name=earth2mip-kernel

Docker Environment#

For the most consistent environment, we suggest using the Modulus docker container that is avaiable on the Nvidia container registry. Launch the Modulus docker container with port mapping as below:

docker run --rm --shm-size=1g --ulimit memlock=-1 --ulimit stack=67108864 --runtime nvidia -p 8888:8888 -it nvcr.io/nvidia/modulus/modulus:23.08

Note

The above command maps the port 8888 inside the docker container to the port 8888 on local machine. This is a requirement needed when using Jupyter Lab to run the notebook examples.

Once inside the container install Earth-2 MIP, using your prefered method.

Optional Dependicies#

Earth-2 MIP has a number of optional dependencies that can be installed depending on the use case.

Models#

Install optional dependencies for Pangu weather:

pip install .[pangu]

Install optional dependencies for Graphcast:

pip install .[graphcast]

Development#

Install development packages for linting/formating/etc:

pip install .[dev]

Install packages for building documentation:

pip install .[docs]