This repo contain experiments investigating how a deconvolution method performs on data relevant for climate simulation output.

The project is developed using pixi. To install, clone the repo and let pixi do the rest.

git clone https://github.com/engeir/gpu-deconv
cd gpu-deconv || exit
pixi install

Pixi is similar to conda, and is able to produce environment files that conda can use to install the exact same project environment. To install this project using conda rather than pixi:

pixi project export conda-environment environment.yml
conda env create -n new-env-name -f environment.yml

Run the scripts as usual as python src/<file>.py, or in some cases pixi tasks are available: pixi run script1.

Some issues may arise when running this code.

• Path to the GPU libraries

  The default path used to look for libraries is not necessarily the correct one, and can be overridden with the LD_LIBRARY_PATH environment variable. When using pixi, this is set in the tasks in the pyproject.toml file. See the fpp-analysis-tools README for how to find your path.

• Plotting on a remote server

  The scripts in this repo do some plotting, and its nice to be able to view them interactively, not just save the image files and inspect those. This need an ssh connection that uses the -X flag (ssh -X user@servername) or that it is set in the ssh configuration at $HOME/.ssh.config:

  Host somecustomname
      ForwardX11 yes
      ForwardX11Timeout 0
      Hostname 1.2.3.4
      User user

  Note that there might be a timeout on the X11 forwarding, so setting this to zero (no timeout) in the configuration is the simplest solution.

  Further, the DISPLAY environment variable must be set once you are logged into the server. Try first echo "$DISPLAY", and if it is empty do export DISPLAY=localhost:10.0. In this repo, this value is set in the .mise.toml file, but this depend on having mise installed to work.