The RL figure can be created with:
python create_rl_figure.py
This loads the runs from rl_data.csv.
Alternatively, reproduce the runs with:
python train.py -m +experiment=sfr-sample,laplace-sample,ensemble-sample,ddpg,mlp ++random_seed=100,69,50,666,54
and then populate the rl_data.csv by fetching the runs from W&B:
python fetch_data_from_wandb.py