AdcircNN - Physics based machine learning model with ADCIRC

AdcircNN is a Python software for physics based machine learning with ADvanced CIRCulation (ADCIRC) and neural networks. AdcircNN is a sister project of Water Coupler and may be integrated into it later.

Getting Started

Dependencies

• Python 3+.
• NumPy
• pyADCIRC, the Python interface of ADCIRC, which requires:
  • ADCIRC shared library : lib*adcpy.so*
  • ADCIRC python interface : pyadcirc*.so

Installing

Following steps are needed to properly install AdcircNN on Linux:

• Install the ADCIRC Python interface, pyADCIRC, and run tests if needed
• Add the path of the shared libraries lib*adcpy.so* and pyadcirc*.so to the LD_LIBRARY_PATH environment variable, and
• Run the following commands to clone this repository, change the directory, and pip-install AdcircNN.

  git clone https://github.com/UT-CHG/adcirc_nn.git
  cd adcirc_nn
  python3 -m pip install .

• You should now be able to import adcirc_nn in Python, if needed.

Running tests

To do.

Using the project

Suppose you want to run an AdcircNN simulation. After installing pyADCIRC and AdcircNN, copy the ADCIRC input files in a single directory. Run AdcircNN as a module with 2 command line arguments as follows.

• Argument 1: Coupling type identifier which is one of {Adn, ndA, AdndA, ndAdn}
• Argument 2: Boundary string ID of the ADCIRC model that is being coupled to the machine learning model, For instance, the Linux/Unix workflow goes as follows.

mkdir sample-sim
cd sample-sim
cp <path_to_ADCIRC_input_files>/fort.* .

python -m adcirc_nn <Coupling type identifier>  <ADCIRC model coupled boundary>

License

AdcircNN is distributed under the BSD 3-Clause "New" or "Revised" license. Note, however, that some of the external dependencies of this software are proprietary/closed-source, which cannot and should not be distributed with this software.