JET DNN

JETDNN: a Python framework for building analytic models relating plasma pedestal heights with multiple engineering parameters, using Deep Neural Networks (DNNs). JETDNN functions cover 3 main areas: data inspection, model building, and visualisation. Created by S. Frankel, J. Simpson and E. Solano.

Installation

To use JETDNN, install it in the console using pip:

$ pip install jetdnn

or

$ pip install JETDNN

Requirements

The packages required to run JETDNN can be found on GitHub here: https://github.com/quasoph/jetdnn/blob/main/requirements.txt.

Functions

JETDNN functions cover three main areas: data inspection, pedestal prediction, and visualisation. Detailed documentation can be found at https://jetdnn.readthedocs.io/en/latest/.

An example workflow can look like:

1. Inspect data with jetdnn.interpret.plot_params() to find engineering parameters strongly correlated to pedestal heights.

2. Train and test a DNN to find a model relating these parameters with pedestal height, using jetdnn.predict.build_and_test_single().

3. Predict pedestal heights using this model from any dataset with jetdnn.predict.predict_single().

4. Visualise model as an analytic equation with functions from jetdnn.visualise.