Simulate Lagrangian motion use generative adversarial network.
https://arxiv.org/abs/1901.03960
Files in repository:
full_dictionary: full trajectory coordinates in r-\theta-z; dictionary, 15 elements, [1100,3] dimension.
input_array: initial coordinates of each 100 coordinates trajectories; dictionary, 165 elements, [6, 3] dimension.
output_array: full coordinates of each 100 coordinates trajectories; dictionary, 165 elements, [100, 3] dimension.
seq2seq_short_GAN.ckpt: trained model parameter after 5000 iterations.
lagrangian_gan.py: training program. Use input_array to generate simulated Lagrangian motion trajectory for adversarial training (compare with output_array).
gan_recovery: model recovery program. Use trained model parameter and full_dictionary to generate full length Lagrangian motion simulations.
sqe2seq_short_training_curve: obtained training curve.