smokeppe_constraints

Implementation of frequentist confidence sets for climate model parameter constraints

Contents

• examples: Files in JSON format instantiating parameters and data domains for constraint examples.
• script: The subdirectory from which methods should be interface by the command line.
• src: Methods to be called indirectly through script.

Use

From the script directory, run the run.py file as a module as follows:

python -m run --input_file <path-to-input-json-file> --output_dir <path-to-output-directory> --savefigs

If desired, set up a default input and output with a config.ini file. For example, create an .ini file in the script subdirectory as follows:

[DEFAULT]
InputFile = /input/directory/evalParameters.json
OutputDir = /output/directory/

See examples/evalParametersTemplate.json for example of the contents of this json file.

Options

Our method calls for several options for the noise model, each requiring bespoke estimation methods.

Gaussian noise model

The Gaussian noise model simply optimizes the closed form likelihood with the scipy implementation of LBFG-S.

Student-t noise model

This model optimizes two values for the student-t approximation, shape $\nu$ and scale $\delta$. For numerical optimization, bounds on the search range will be included in the JSON example configuration file where the first set of bounds will correspond to $\delta_{MLE}$ and the second set of bounds will correspond to $\nu$. It is required mathematically that $\nu>2$ and that $\delta>0$.