ML-RT stands for Machine Learning and Radiative Transfer. It is a project aimed at examining the possibility of using machine learning techniques to emulate radiative transfer simulation results.
We assume your system is equipped with the following dependencies:
- Python 3.8 or newer
- bash
- wget
- unzip
- md5sum (optional)
On Debian or Debian-derivatives, e.g. Ubuntu, the required packages should be part of the base installation but can be installed using the default package manager if necessary with the following command:
sudo apt install wget unzip md5sumFurthermore, the following Python packages are needed:
- pytorch
- numpy (1.20.0 or newer)
- numba (used in the Soft DTW implementation)
- matplotlib
- seaborn
- gdown (to download pre-trained models)
The Python dependencies can be installed with pip like so:
pip3 install -r requirements.txtIn Anaconda (or Miniconda) environments the requirements can be installed like so:
conda config --add channels conda-forge
conda install --yes --file requirements_conda.txttba.
To run pretrained models with custom input, you will need to download the pretrained models first. To do this,
navigate to the paper_data directory via cd ./paper_data and run the pretrained_models.sh script:
bash ./pretrained_models.shThis script will download and extract all pretrained models. To run inference on them, navigate to the src folder,
which contains the run scripts, including inference.py:
cd ./src/
python3 inference.pyBy default, the script will create a plot using the default set parameters in the script. It can easily be adapted to
run inference with custom parameters. Furthermore, one can select what models to use, measure inference time, and more.
The main function at the bottom of inference.py features a sample implementation.
tba.
Feedback and questions are welcome. Please get in touch.