phlower.nn.IsoGCN

class phlower.nn.IsoGCN(nodes, isoam_names, propagations, use_self_network, self_network_activations=None, self_network_dropouts=None, self_network_bias=False, use_coefficient=False, coefficient_activations=None, coefficient_dropouts=None, coefficient_bias=False, mul_order='ah_w', to_symmetric=False, use_neumann=False, neumann_factor=1.0, neumann_input_name=None, inversed_moment_name='')

Bases: IPhlowerCoreModule, Module

IsoGCN is a neural network module that performs an spatial pifferential operation on the input tensor. Ref: https://arxiv.org/abs/2005.06316

Parameters:

• nodes (list[int]) – List of feature dimension sizes (The last value of tensor shape).

• isoam_names (list[str]) – List of names of the isoam tensors.

• (list[IsoGCNPropagationType]) (propagations) – List of propagations to apply to the input tensor.

• use_self_network (bool) – Whether to use the self network.

• self_network_activations (list[str] | None (optional)) – List of activation functions to apply to the output of the self-network. Defaults to None.

• self_network_dropouts (list[float] | None (optional)) – List of dropout rates to apply to the output of the self-network. Defaults to None.

• self_network_bias (bool | None (optional)) – Whether to use bias in the self-network. Defaults to False.

• use_coefficient (bool) – Whether to use the coefficient network.

• coefficient_activations (list[str] | None (optional)) – List of activation functions to apply to the output of the coefficient network. Defaults to None.

• coefficient_dropouts (list[float] | None (optional)) – List of dropout rates to apply to the output of the coefficient network. Defaults to None.

• coefficient_bias (bool | None (optional)) – Whether to use bias in the coefficient network. Defaults to False.

• (Literal["ah_w" (mul_order) – Multiplication order. Defaults to “ah_w”.

• "a_hw"]) – Multiplication order. Defaults to “ah_w”.

• to_symmetric (bool) – Whether to make the output symmetric. Defaults to False.

• use_neumann (bool) – Whether to use the Neumann layer. Defaults to False.

• neumann_factor (float) – Factor to multiply the Neumann layer. Defaults to 1.0.

• neumann_input_name (str | None (optional)) – Name of the input tensor for the Neumann layer. Defaults to None.

• inversed_moment_name (str) – Name of the inversed moment tensor. Defaults to “”.

• propagations (list[IsoGCNPropagationType])

• mul_order (Literal['ah_w', 'a_hw'])

Examples

>>> # Create an IsoGCN instance for laplacian operation with neumann condition
>>> isogcn = IsoGCN(
...     nodes=[10, 20, 30],                    # Feature dimensions
...     isoam_names=["isoam_x", "isoam_y", "isoam_z"],
...     propagations=[                          # Types of propagation
...         IsoGCNPropagationType.convolution,
...         IsoGCNPropagationType.contraction
...     ],
...     # Self network configuration
...     use_self_network=True,
...     self_network_activations=["relu", "relu"],
...     self_network_dropouts=[0.1, 0.1],
...     self_network_bias=True,
...     # Coefficient network configuration
...     use_coefficient=True,
...     coefficient_activations=["relu", "relu"],
...     coefficient_dropouts=[0.1, 0.1],
...     coefficient_bias=True,
...     # Additional settings
...     mul_order="ah_w",
...     to_symmetric=True,
...     use_neumann=True,
...     neumann_factor=1.0,
...     neumann_input_name="neumann_input",
...     inversed_moment_name="inversed_moment"
... )
>>> # Forward pass
>>> output = isogcn(data)

Methods

forward(data, *, field_data, **kwards)	forward function which overload torch.nn.Module
from_setting(setting)	Create IsoGCN from IsoGCNSetting instance
get_nn_name()	Return neural network name

Attributes

T_destination
call_super_init
dump_patches
training

forward(data, *, field_data, **kwards)

forward function which overload torch.nn.Module

Parameters:

• data (IPhlowerTensorCollections) – IPhlowerTensorCollections data which receives from predecessors

• field_data (ISimulationField | None) – ISimulationField | None Constant information through training or prediction

Returns:

Tensor object

Return type:

PhlowerTensor

classmethod from_setting(setting)

Create IsoGCN from IsoGCNSetting instance

Parameters:

setting (GCNSetting) – setting object for IsoGCN

Returns:

IsoGCN object

Return type:

IsoGCN

classmethod get_nn_name()

Return neural network name

Returns:

name

Return type:

str