nidn.training.model package
Submodules
nidn.training.model.init_network module
- nidn.training.model.init_network.init_network(run_cfg: DotMap)
Network architecture. Note that the dimensionality of the first linear layer must match the output of the encoding chosen in the config file. :returns: Initialized model :rtype: torch model
nidn.training.model.model_to_eps_grid module
- nidn.training.model.model_to_eps_grid.model_to_eps_grid(model, run_cfg: DotMap)
Computes a 4D grid of epsilons for the given model.
- Parameters
model (torch.model) – Trained neural network model. Should map one 4D input to a [real,imag] epsilon.
run_cfg (DotMap) – Configuration for the run.
- Returns
Resulting 4D [real,imag] epsilon grid and tensor of the material_ids (None for regression)
- Return type
torch.tensor, torch.tensor