hipnn module

Full Documentation for hippynn.networks.hipnn module. Click here for a summary page.

Implementation of HIPNN.

class Hipnn(n_features, n_sensitivities, dist_soft_min, dist_soft_max, dist_hard_max, n_atom_layers, n_interaction_layers=None, possible_species=None, n_input_features=None, sensitivity_type='inverse', resnet=True, activation=<class 'torch.nn.modules.activation.Softplus'>, cusp_reg=None)

Bases: Module

Hipnn Main Module

forward(features, pair_first, pair_second, pair_dist)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

regularization_params()

property interaction_layers

property sensitivity_layers

class HipnnQuad(*args, cusp_reg=1e-06, **kwargs)

Bases: HipnnVec

HIP-NN-TS with l=2

class HipnnVec(*args, cusp_reg=1e-06, **kwargs)

Bases: Hipnn

HIP-NN-TS with l=1

forward(features, pair_first, pair_second, pair_dist, pair_coord)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

compute_hipnn_e0(encoder, Z_Data, en_data, peratom=False, fit_dtype=torch.float64)

Parameters:

• encoder – encoder of species to features (one-hot representation, probably)

• Z_Data – species data

• en_data – energy data

• peratom – whether energy is per-atom or total

Returns:

energy per species as shape (n_features_encoded, 1)