Distribution Statement A: Distribution Statement A. Approved for public release, distribution is unlimited.

Introduction

CodeJeNN is a simple neural network generator for c++ that can robustly ingest a trained NN to be used on the fly for inference in target computational physics and fluid dynamics software. This abstracts away the need for using third party libraries which are often cumbersome and would require shipping large data onto main memory to utilize inference. Instead, the NN is inlined and localized to the users machine for inference. This will assure scalability and allow for certain optimizations that ML has promised to numerical solvers: faster constiutive laws, accurate interpolation functions, etc. CodeJeNN works by converting a trained neural net stored in a .h5 or a .keras file into a c++ header file that can be used in the users code to predict (perform inference).

Directory Contents

CodeJeNN/
    └── src/
            └── api-core/
            └── bin/
            └── dump_model/
            └── clean.sh
            └── generate.sh
            └── readme.md
    └── tutorials/
            └── 01_simple_mlp/
            └── 02_cnn_1d/
            └── 03_cnn_2d/
            └── 04_cnn_3d/
            └── 05_advanced_mlp/
            └── hdf5_file_breakdown.md
            └── supported_layers.md
    └── citation.cff
    └── license.txt
    └── logo.png
    └── readme.md
    └── requirements.txt

Starting Point

! Python 3.11 is required due to compatibility with the latest version of tensorflow (NOTE: Keras is used through tensorflow so only installing tensorflow is necessary).

! The recommended installer is apt for Linux users and brew for macOS users.

! You ONLY need the src directory, all other files are just auxillary but still very useful.

1. First open up a terminal/shell session and clone this repo into the home "~/" directory:

   git clone https://github.com/jarcities/CodeJeNN.git ~/codejenn
   cd ~/codejenn

   or where ever you choose:

   git clone https://github.com/jarcities/CodeJeNN.git
   cd codejenn

2. You do not need to create a virtual environment, but it is best to use one. This allows all dependent packages to be in one spot.

   The first way is by using conda which you can install from Install Miniconda (official site). Then in your terminal/shell, create a conda environment and activate it.

   conda create -n codejenn python=3.11
   conda activate codejenn

   OR

   The second way is to use a python environment by installing python 3.11 using sudo apt install python3.11 or brew install python@3.11. Then in your home directory create a python_environments directory and create an environtment in there and activate it.

   mkdir ~/python_environments/
   cd ~/python_environments/
   python3.11 -m venv codejenn
   source codejenn/bin/activate

   Where codejenn is the name of the environment.

3. Next install the necessary libraries which are common in most deep learning codes already.

   pip install -r requirements.txt

4. From here, cd into src and carry on with the README.md file in there.

Citation

# WILL ADD SOON