Fig. 1: dead_zone_plots.py (dead_zone_preparation.py for preparation, ./dead_zone/ for files needed)
Fig. 2: all_plots_main_paper.py (files needed: "./testing/human_vs_agnes_statistics.csv", "./testing/stats_comparison.csv", "return_trainings_stats.csv", "return_trainings_long_stats.csv")
Fig.13: return_plots_appendix_replay_goal.py (files needed: return_start_from_0_stats.csv, return_replay_buffer_stats.csv)
    (return_start_from_0_stats.csv with smoothed means and standard deviation of return of training without pre-training
    return_replay_buffer_stats.csv with smoothed means and standard deviation of return of training pre-training on lower goals, with or without deleting the replay buffer when increasing the goal)
Fig.14: return_plots_appendix_pre_virtual.py (files needed: return_training_pretrained_virtual_stats.csv, return_replay_buffer_stats_without_replay.csv)
   (return_training_pretrained_virtual_stats.csv with smoothed means and standard deviation of return of training pre-training on virtual testbed
    return_replay_buffer_stats_without_replay.csv with smoothed means and standard deviation of return of training pre-training on lower goals, deleting the replay buffer when increasing the goal)

./return.../: include original csv files downloaded from tensorboard with the return during training
Using return_..._preparation.py files, these are then merged and prepared for plots (by smoothing, and determining the mean and standard deviation)

./testing/: includes test runs appearing in the paper (names: {timestamp of training run}_{training step at which model is tested}
