Plasim routines

Intermediate complexity climate model. Ideal for method development: allows very long simulations

Folder structure

• Current setup (all functions relate to Learn2_new.py):

  • Learn2_new.py is responsible for training the CNN. It includes a range of different routines for training, testing, and analyzing the CNN. Please consult the tutorial.ipynb for use cases and the documentation provided in Learn2_new.py. It helps to look at example_config.json to understand the structure of the routines and the kwargs used.
  • hyperparameter_optimization.py contains the routine which optimizes hyperparameters for the CNN using optuna - Bayesian parameter optimization framework. The script hyperparameter_optimization.py also takes trainer object from Learn2_new.py.
  • network_inspection_tools.py contains a set of routines which allow to inspect the CNN and its performance.
  • resampling-sensitivity.py is a routine which allows to test the sensitivity of the CNN to the resampling of the data.
  • inheritance-template.py which shows how to efficiently inherit from the CNN class.
  • import_config.py is a routine which allows to automatically update the config.json file of the current run with the kwargs of a previous run.

• Plasim data analysis:

  • Plasim_France.ipynb and Plasim_Scandinavia.ipynb are notebooks which contain the analysis of the Plasim data for France and Scandinavia heatwaves. It is a good example of how to work with the data, and how to use the Plasim class. It contains statistical analysis of the temperature time series and geopotential teleconnection patterns.

• Older routines (support discontinued):

  • Learn2.py is the older version of Learn2_new.py.
  • History.py allows plotting the history of the training of the CNN. It is used for the outputs of Learn2.py.
  • Recalc_Tau_Metrics.py/ Plot_Tau_Metrics.py are routines which allow to recalculate the metrics of the CNN for a given tau value. It is used for the outputs of Learn2.py.
  • Recalc_Comm_Faster.ipynb is a notebook which plots the committor function of the CNN for a given tau value. It is used for the outputs of Learn2.py.
  • Committor_LONG-smoothness.ipynb is a notebook which investigates the smoothness of a committor learned by CNN as a function of tau. It is used for the outputs of Learn2.py.