environment

Environment Setup

This project is developed in Python 3.9. You can choose to use a conda/virtual environment, or use the provided docker image (available in Dockerhub.

Conda environment

There are multiple options for which environment to use depending on your use case. In the following commands, you can use a different environment name, by changing string that comes after the -n flag in the conda env create ... lines

Training (and evaluating)

On GPU

conda env create -f env_train_gpu.yaml -n MY_ENV_NAME
conda activate MY_ENV_NAME  # activate the environment called <MY_ENV_NAME>

On CPU

conda env create -f env_train_cpu.yaml -n MY_ENV_NAME
conda activate MY_ENV_NAME  

Only evaluate and do inference

The following environment file is for inference only, i.e. if you want to evaluate (or predict with) a model that has already been trained.

conda env create -f env_evaluation.yaml -n climateml-eval
conda activate climateml-eval  

Note for jupyter notebooks:

You need to choose the above environment (e.g. climateml-eval) as kernel of the jupyter notebook. If the environment doesn't show up in the list of possible kernels, please do

python -m ipykernel install --user --name climateml-eval   # change aibedo with whatever environment name you use 

Then, please refresh the notebook page.

Docker container

See the build-push-docker-image script for how to build and push the docker image to {Dockerhub, Nautilus registry}.