kth-flowai / identifying-regions-of-importance-in-wall-bounded-turbulence-through-explainable-deep-learning

Code for reproducing the article: Identifying regions of importance in wall-bounded turbulence through explainable deep learning.

In this repository an example of the code is provided.

Some modifications have been applied to the original code to make it run under limited resources and with a smaller database.

To create the data folder, please execute: $ ./createdata.sh

In order to reproduce the results of the paper make the following changes:

• Simulation_3d:

  • main_segment.py:

    line 14: start = 1000
    line 15: end = 9999

  • main_CNN.py:

    line 21: CNN.define_model(shp=(shpy,shpz,shpx,3),learat=1e-2) line 22: CNN.train_model(1000,7000,delta_t=80,delta_e=200,max_epoch=2e2,
    batch_size=8,down_y=dy,down_z=dz,down_x=dx)

  • main_evCNN.py:

    line 14: end = 9999

  • main_evCNN_error.py:

    line 13: end = 9999

  • main_shap.py

    line 12: end = 9999

  • readshap.py

    line 12: end = 9999

  • shap_config.py:

    line 66: shap_values = explainer.shap_values(zshap,nsamples="auto")[0][0]

• Experiment_2d:

  • main_segment.py:

    line 15: end = 5978

  • main_CNN.py:

    line 22: CNN.train_model(1,5978,delta_t=5978,delta_e=50,max_epoch=1.3e3,
    batch_size=128,down_y=dy,down_x=dx)

  • shap_config.py:

    line 88: shap_values = explainer.shap_values(zshap,nsamples="auto")[0][0]

In addition, the complete database is required.

• Simulation_3d: from field 1000 to 9999 total of 9000 fields
• Experiment_2d: from field 1_0 to 0_5978 total of 6100 fields

The repository contains the minimum data to run all the simulations. We recommend running them using the run file.

In case of running the files separately, some issues may occur due to the lack of data which is generated by a previous file.

For any trouble or doubt, please contact us:

Andres Cremades Botella: [email protected]