CLEAR AI pipeline

To use this pipeline, set up a python environment with the packages from requirements.txt.

The pipeline consists of several consecutive steps

1. converting trajectories into images of the trajectories
2. train the neural network using the BYOL framework
3. calculate the embeddings of the images

Convert the trajectories into images

For this, we use analysis/prepare_2layer_data.py. We supply the script with the trajectories output file from the FAIRSEA code (e.g. reduced_data_20241022_1105.pkl).

Train the CNN

We train the neural network with analysis/train_2layer_model.py. It is important to supply a configfile that follows the conventions from analysis/example_config.toml

Calculate the embeddings