search

PPPLDeepLearning / plasma-python

PPPL deep learning disruption prediction package
http://tigress-web.princeton.edu/~alexeys/docs-web/html/
79 stars 43 forks source link

Make signals positive #21

Closed ghost closed 6 years ago

ghost commented 6 years ago

Added a is_strictly_positive flag to the Signal object, and forced the normalizer to apply a positivity constraint to negative data to be 0 for strictly positive signals.

Example: plasma density. Physical values should be strictly positive. Several shots have negative plasma density:

File: 79569.txt Mean: -2.71116351086e+19
File: 74629.txt Mean: -4.08420997895e+20
File: 74822.txt Mean: -4.38447577376e+22
File: 74836.txt Mean: -4.96838852943e+22
File: 74846.txt Mean: -1.89555301098e+23
File: 76419.txt Mean: -1.9323283152e+23
File: 74584.txt Mean: -2.20456379055e+21
File: 74870.txt Mean: -9.72736715272e+22
File: 74827.txt Mean: -2.00043820891e+28
File: 83641.txt Mean: -7.90047309436e+19
File: 76197.txt Mean: -3.41959747107e+19
File: 75937.txt Mean: -9.1022051383e+20
File: 76418.txt Mean: -2.124980498e+19
File: 74849.txt Mean: -2.34416121644e+23
File: 74628.txt Mean: -5.83265091709e+19
File: 77019.txt Mean: -5.02673980466e+19
File: 76300.txt Mean: -4.44121046833e+19
ASvyatkovskiy commented 6 years ago

Example shot where plasma density becomes negative (JET, shot #74827) image

ASvyatkovskiy commented 6 years ago

@soccerturtle7 Good work, Joe! Please provide summary of normalization constants (means and standard deviations per signal) for various types of normalization for entire JET dataset

ASvyatkovskiy commented 6 years ago

Continued in PR https://github.com/PPPLDeepLearning/plasma-python/pull/22