Datasets and scripts for 'Early warning signals of tipping points: A deep learning approach to direction and timing'

Files included: categorical_model_train.R categorical_model_test.R ttt_model_train.R ttt_model_test.R saddle_node_cat_training.RData pitchfork_cat_training.RData transcritical_cat_training.RData saddle_node_cat_testing.RData pitchfork_cat_testing.RData transcritical_cat_testing.RData saddle_node_ttt_training.RData pitchfork_ttt_training.RData transcritical_ttt_training.RData saddle_node_ttt_testing.RData pitchfork_ttt_testing.RData transcritical_ttt_testing.RData best_categorical_model.pt best_ttt_model.pt 4 .R scripts for training and testing models of the same architecture as in the paper, either ‘categorical’ or ‘ttt’ (time to tip) and ‘training’ and ‘testing’. Both the ‘torch’ and ‘coro’ packages are required. 12 .RData files containing ‘training’ and ‘testing’ datasets for each normal form (‘saddle_node’, ‘pitchfork’ and ‘transcritical’) for both the ‘cat’ (categorical) and ‘ttt’ models. The testing datasets for the categorical model also include the change in bifurcation parameter r over the 1500 time steps, so further analysis can be carried out if the user desires, and in the time-to-tip model, the start and end values of bifurcation parameter r. 2 .pt files that can be loaded into the testing scripts that represent the categorical and ttt models used in the original analyses. Running the training scripts produces a timestamped .pt model e.g. ‘best_categorical_model_YYYYMMDD_HHMMSS.pt’ which can be used in the testing scripts by changing the ‘model_path’ in the testing scripts. By default it will use the .pt files included. Scripts and analysis was carried out using R ver. 4.4.1 and torch ver. 0.15.0 Scripts to generate further time series for training can be found at: https://github.com/caboulton/AdvanTip_Public