main.py

This is the main file to train the ICNN-based material model.

The arguments to run the file are:

• <fem_material> - can be any one of the following: NeoHookean, Isihara, HainesWilson, GentThomas, ArrudaBoyce, Ogden, Anisotropy45, Anisotropy60, Holzapfel
• <noise_material> - noise conditioning of the data (can be low or high)

The individual components of the main file are the following:

• Reads the command line arguments <fem_material> and <noise_material> and loads the datasets accordingly.
• Initializes ICNN model and assigns randomly sampled weights and biases to it (sampled via xavier_uniform)
• Trains a number of ICNNs (equal to ensemble_size defined in config.py) each with randomly initialized weights.
• Saves each model to the output directory defined in config.py.
• Once all models are trained evaluate_icnn() is called to evaluate each ICNN in the ensemble and plot the performance against the ground truth model along six deformation paths.