Emulating Spatio-Temporal Realizations of Three-Dimensional Isotropic Turbulence via Deep Sequence Learning Models

This is an implementation of Emulating Spatio-Temporal Realizations of Three-Dimensional Isotropic Turbulence via Deep Sequence Learning Models

Requirements

• see requirements.txt

Instruction

• Global hyperparameters are configured in config.yml
• Hyperparameters can be found at process_control() in utils.py

Examples

• Train vqvae, compression scale 1, regularization parameter

  python train_vqvae.py --data_name Turb --model_name vqvae --control_name 1_chs-1_exact-physcis_0.1_0

• Train Conv-LSTM model, compression scale 2, regularization parameter , 3 time steps for training and 3 time steps for testing, with teacher-forcing

  python train_convlstm.py --data_name Turb --model_name convlstm --control_name 2_chs-1_exact-physcis_0.1_0.001_3-3_0

• Train Conv-Transformer model, compression scale 3, regularization parameter , 3 time steps for training and 3 time steps for testing, with cyclic training

  python train_transformer_cyclic.py --data_name Turb --model_name transformer --control_name 3_chs-1_exact-physcis_0.1_0.001_3-3_0_1

Results

• Predictions of velocity field using Conv-LSTM models.

• Predictions of velocity field using Conv-Transformer models.

Acknowledgement

Mohammadreza Momenifar
Enmao Diao
Vahid Tarokh
Andrew D. Bragg