Regularization extraction for real-time plasma tomography at JET [paper] [poster]
D. R. Ferreira, D. D. Carvalho, P. J. Carvalho, H. Fernandes, and JET Contributors
45th EPS Conference on Plasma Physics, Prague, Czech Republic, July 2-6, 2018

• geom.txt and kb5_los.txt contain the geometry for the vessel and the KB5 lines of sight, respectively.

• geom.py contains some simple routines to read those geometry files.

• tomo_kb5_reliable.hdf contais the reconstructions that were used for data fiting; bolo_kb5_reliable.hdf contains the full KB5 signals for those same pulses.

• fit_M.py fits the M matrix using the full dataset; fit_validate_M.py fits M using 90% for training and 10% for validation.

• train.png and valid.png contain the computation graphs generated by Theano for the train and validation functions, respectively.

• train.log contains the loss and validation loss recorded during training.

• plot_train.py plots the loss and validation loss during (or after) training.

• When training finishes, M will be saved to M.npy.

• plot_M.py plots the regularization patterns found in the columns of M.

• metrics.py calculates SSIM, PNSR and NRMSE on the validation set, after M has been trained.

• create_movie.py generates a full-pulse reconstruction video for shot 92213.

• 92213_47.00_54.50.mp4 contains the generated movie.