README.md

RDCNN

RDCNN predicts the result of reaction-diffusion system based on convolutional neural network (CNN)

User guide

User can run the code following the steps below.

1. FEM_Data_Geneartion (C++)

• Description: this code is used to generate the dataset using FEM.

• Input:

  • mesh_21_21.vtk (Quadrilateral mesh for domain geometry)

• Output:

  • geometry_X_Y_input.txt (the input data storing boundary condition corresponding to Xth geometry and Yth parameter settings)
  • mesh_X.txt (the output data storing concentration results)
  • dataset_DKTGeo.txt (the library stores the parameter setting of each sample)

• To compile: (requires Eigen)

  >> make

• To run:

  Create a folder (eg. "\data") first and then create another 3 folders in this folder ("\data\input", "\data\output" and "\data\parametric") to store input data, output data and visualization results separately.

  >> ./rdfem -t <val_t> -s <val_s> -g <val_g> -o <output_path>

  output_path is the output path for data generation

  Example:

  >> ./rdfem -t 100 -s 500 -g 21 -o ../data/

2. txt_hdf5.ipynb

• Description: This code is used to transform data format from TXT to H5
• Input:
  • ./data (Dataset folder)
• Output:
  • X.h5 (Dataset stored in h5 file)
• To run： User can open the code in jupyter notebook follow the comments to run the code.

3. Main_data_rdcnn2larger.ipynb

• Description: this code is used to train CNN model and predict concentration results for the specific reaction diffusion system.

• To run： User can open the code in jupyter notebook follow the comments to run the code.