pyg_spectral.nn.models

class BaseNN(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: Module

Base NN structure with MLP before and after convolution layers.

Parameters:

• conv (str) – Name of pyg_spectral.nn.conv module.

• num_hops (int, default: 0) – Total number of conv hops.

• in_channels (int | None, default: None) – Size of each input sample.

• hidden_channels (int | None, default: None) – Size of each hidden sample.

• out_channels (int | None, default: None) – Size of each output sample.

• in_layers (int | None, default: None) – Number of MLP layers before conv.

• out_layers (int | None, default: None) – Number of MLP layers after conv.

• dropout_lin (float | List[float], default: 0.0) – Dropout probability for both MLPs.

• dropout_conv (float, default: 0.0) – Dropout probability before conv.

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• lib_conv – Parent module library other than pyg_spectral.nn.conv.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

supports_edge_weight: Final[bool] = False

supports_edge_attr: Final[bool] = False

supports_batch: Final[bool] = False

supports_norm_batch: Final[bool]

init_channel_list(conv: str, in_channels: int, hidden_channels: int, out_channels: int, **kwargs) → List[int]

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

_set_conv_attr(key: str) → Callable

reset_cache()

reset_parameters()

get_optimizer(dct)

preprocess(x: Tensor, edge_index: Tensor | SparseTensor) → Any

Preprocessing step that not counted in forward() overhead. Here mainly transforming graph adjacency to actual propagation matrix.

convolute(x: Tensor, edge_index: Tensor | SparseTensor, batch: Tensor | None = None, batch_size: int | None = None) → Tensor

Decoupled propagation step for calling the convolutional module.

forward(x: Tensor, edge_index: Tensor | SparseTensor, batch: Tensor | None = None, batch_size: int | None = None) → Tensor

Parameters:

• x (Tensor) – from torch_geometric.data.Data

• edge_index (Tensor | SparseTensor) – from torch_geometric.data.Data

• batch (Tensor | None, default: None) – The batch vector \(\mathbf{b} \in {\{ 0, \ldots, B-1\}}^N\), which assigns each element to a specific example. Only needs to be passed in case the underlying normalization layers require the batch information.

• batch_size (int | None, default: None) – The number of examples \(B\). Automatically calculated if not given. Only needs to be passed in case the underlying normalization layers require the batch information.

class BaseNNCompose(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNN

Base NN structure with multiple conv channels.

Parameters:

• combine (str) – How to combine different channels of convs. (sum, sum_weighted, or cat).

• conv (str) – Name of pyg_spectral.nn.conv module.

• num_hops (int, default: 0) – Total number of conv hops.

• in_channels (int | None, default: None) – Size of each input sample.

• hidden_channels (int | None, default: None) – Size of each hidden sample.

• out_channels (int | None, default: None) – Size of each output sample.

• in_layers (int | None, default: None) – Number of MLP layers before conv.

• out_layers (int | None, default: None) – Number of MLP layers after conv.

• dropout_lin (float | List[float], default: 0.0) – Dropout probability for both MLPs.

• dropout_conv (float, default: 0.0) – Dropout probability before conv.

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• lib_conv – Parent module library other than pyg_spectral.nn.conv.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_channel_list(conv: str, in_channels: int, hidden_channels: int, out_channels: int, **kwargs) → List[int]

Variables:

channel_list – width for each conv channel

_set_conv_attr(key: str) → List[Callable]

reset_cache()

reset_parameters()

get_optimizer(dct)

preprocess(x: Tensor, edge_index: Tensor | SparseTensor) → Any

Preprocessing step that not counted in forward() overhead. Here mainly transforming graph adjacency to actual propagation matrix.

convolute(x: Tensor, edge_index: Tensor | SparseTensor, batch: Tensor | None = None, batch_size: int | None = None) → Tensor

Decoupled propagation step for calling the convolutional module.

class Iterative(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNN

Iterative structure with matrix transformation each hop of propagation.

Parameters:

• bias (Optional[bool]) – whether learn an additive bias in conv.

• weight_initializer (Optional[str]) – The initializer for the weight matrix ("glorot", "uniform", "kaiming_uniform", or None).

• bias_initializer (Optional[str]) – The initializer for the bias vector ("zeros" or None).

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

class IterativeCompose(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNNCompose

Iterative structure with matrix transformation each hop of propagation.

Parameters:

• bias (Optional[bool]) – whether learn an additive bias in conv.

• weight_initializer (Optional[str]) – The initializer for the weight matrix ("glorot", "uniform", "kaiming_uniform", or None).

• bias_initializer (Optional[str]) – The initializer for the bias vector ("zeros" or None).

• combine – How to combine different channels of convs. (sum, sum_weighted, or cat).

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

class DecoupledFixed(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNN

Decoupled structure without matrix transformation during propagation. Fixed scalar propagation parameters.

Note

Apply conv every forward() call. Not to be mixed with Precomputed models.

Parameters:

• theta_scheme (str) – Method to generate decoupled parameters.

• theta_param (float, optional) – Hyperparameter for the scheme.

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

class DecoupledVar(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNN

Decoupled structure without matrix transformation during propagation. Learnable scalar propagation parameters.

Note

Apply conv every forward() call. Not to be mixed with Precomputed models.

Parameters:

• theta_scheme (str) – Method to generate decoupled parameters.

• theta_param (float, optional) – Hyperparameter for the scheme.

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

reset_parameters()

class DecoupledFixedCompose(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNNCompose

Decoupled structure without matrix transformation during propagation. Fixed scalar propagation parameters.

Parameters:

• theta_scheme (List[str]) – Method to generate decoupled parameters.

• theta_param (List[float], optional) – Hyperparameter for the scheme.

• combine – How to combine different channels of convs. (sum, sum_weighted, or cat).

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

class DecoupledVarCompose(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNNCompose

Decoupled structure without matrix transformation during propagation. Learnable scalar propagation parameters.

Parameters:

• theta_scheme (List[str]) – Method to generate decoupled parameters.

• theta_param (List[float], optional) – Hyperparameter for the scheme.

• combine – How to combine different channels of convs. (sum, sum_weighted, or cat).

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

reset_parameters()

class PrecomputedFixed(in_layers: int | None = None, **kwargs)

Bases: DecoupledFixed

Decoupled structure with precomputation separating propagation from transformation. Fixed scalar propagation parameters and accumulating precompute results.

Note

Only apply propagation in convolute(). Not to be mixed with Decoupled models.

Parameters:

• theta_scheme (str) – Method to generate decoupled parameters.

• theta_param (Optional[float]) – Hyperparameter for the scheme.

• conv – args for BaseNN

• num_hops – args for BaseNN

• in_channels – args for BaseNN

• hidden_channels – args for BaseNN

• out_channels – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers – args for BaseNN

• dropout_lin – args for BaseNN

• dropout_conv – args for BaseNN

• lib_conv – args for BaseNN

• act – args for torch_geometric.nn.models.MLP.

• act_first – args for torch_geometric.nn.models.MLP.

• act_kwargs – args for torch_geometric.nn.models.MLP.

• norm – args for torch_geometric.nn.models.MLP.

• norm_kwargs – args for torch_geometric.nn.models.MLP.

• plain_last – args for torch_geometric.nn.models.MLP.

• bias – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

convolute(x: Tensor, edge_index: Tensor | SparseTensor) → Tensor

Decoupled propagation step for calling the convolutional module. Requires no variable transformation in conv.forward().

Returns:

embed (Tensor) – Precomputed node embeddings.

forward(x: Tensor, batch: Tensor | None = None, batch_size: int | None = None) → Tensor

Parameters:

• x (Tensor) – the output embed from convolute().

• batch (Tensor | None, default: None) – Args for BaseNN

• batch_size (int | None, default: None) – Args for BaseNN

class PrecomputedVar(in_layers: int | None = None, **kwargs)

Bases: DecoupledVar

Decoupled structure with precomputation separating propagation from transformation. Learnable scalar propagation parameters and storing all intermediate precompute results.

Note

Only apply propagation in convolute(). Not to be mixed with Decoupled models.

Parameters:

• theta_scheme (str) – Method to generate decoupled parameters.

• theta_param (Optional[float]) – Hyperparameter for the scheme.

• conv – args for BaseNN

• num_hops – args for BaseNN

• in_channels – args for BaseNN

• hidden_channels – args for BaseNN

• out_channels – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers – args for BaseNN

• dropout_lin – args for BaseNN

• dropout_conv – args for BaseNN

• lib_conv – args for BaseNN

• act – args for torch_geometric.nn.models.MLP.

• act_first – args for torch_geometric.nn.models.MLP.

• act_kwargs – args for torch_geometric.nn.models.MLP.

• norm – args for torch_geometric.nn.models.MLP.

• norm_kwargs – args for torch_geometric.nn.models.MLP.

• plain_last – args for torch_geometric.nn.models.MLP.

• bias – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

convolute(x: Tensor, edge_index: Tensor | SparseTensor) → list

Decoupled propagation step for calling the convolutional module. _forward() should not contain derivable computations.

Returns:

embed – List of precomputed node embeddings of each hop. Each shape is \((|\mathcal{V}|, F, |convs|+1)\).

forward(xs: Tensor, batch: Tensor | None = None, batch_size: int | None = None) → Tensor

Parameters:

• x – the output embed from convolute().

• batch (Tensor | None, default: None) – Args for BaseNN

• batch_size (int | None, default: None) – Args for BaseNN

class PrecomputedFixedCompose(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: DecoupledFixedCompose

Decoupled structure with precomputation separating propagation from transformation. Fixed scalar propagation parameters and accumulating precompute results.

Parameters:

• theta_scheme (List[str]) – Method to generate decoupled parameters.

• theta_param (List[float], optional) – Hyperparameter for the scheme.

• combine – How to combine different channels of convs. (sum, sum_weighted, or cat).

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

convolute(x: Tensor, edge_index: Tensor | SparseTensor) → Tensor

Decoupled propagation step for calling the convolutional module. Requires no variable transformation in conv.forward().

Returns:

embed (Tensor) – Precomputed node embeddings. (shape: \((|\mathcal{V}|, F, Q)\))

forward(x: Tensor, batch: Tensor | None = None, batch_size: int | None = None) → Tensor

Parameters:

• x (Tensor) – the output embed from convolute().

• batch (Tensor | None, default: None) – Args for BaseNN

• batch_size (int | None, default: None) – Args for BaseNN

class PrecomputedVarCompose(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: DecoupledVarCompose

Decoupled structure with precomputation separating propagation from transformation. Learnable scalar propagation parameters and storing all intermediate precompute results.

Parameters:

• theta_scheme (List[str]) – Method to generate decoupled parameters.

• theta_param (List[float], optional) – Hyperparameter for the scheme.

• combine – How to combine different channels of convs. (sum, sum_weighted, or cat).

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

convolute(x: Tensor, edge_index: Tensor | SparseTensor) → Tensor

Decoupled propagation step for calling the convolutional module. Requires no variable transformation in conv.forward().

Returns:

embed (Tensor) – List of precomputed node embeddings of each hop. Shape: \((|\mathcal{V}|, F, Q, |convs|+1)\).

forward(xs: Tensor, batch: Tensor | None = None, batch_size: int | None = None) → Tensor

Parameters:

• x – the output embed from convolute().

• batch (Tensor | None, default: None) – Args for BaseNN

• batch_size (int | None, default: None) – Args for BaseNN

class AdaGNN(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNN

Decoupled structure with diag transformation each hop of propagation.

Paper:

AdaGNN: Graph Neural Networks with Adaptive Frequency Response Filter

Ref:

https://github.com/yushundong/AdaGNN

Parameters:

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

reset_parameters()

class ACMGNN(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNN

Iterative structure for ACM conv.

Paper:

Revisiting Heterophily For Graph Neural Networks

Paper:

Complete the Missing Half: Augmenting Aggregation Filtering with Diversification for Graph Convolutional Networks

Ref:

https://github.com/SitaoLuan/ACM-GNN

Parameters:

• theta_scheme (str) – Channel list. “FBGNN”=”low-high”, “ACMGNN”=”low-high-id”, (“ACMGNN+”=”low-high-id-struct”, not implemented).

• weight_initializer (str, optional) – The initializer for the weight.

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

class ACMGNNDec(conv: str, num_hops: int = 0, in_channels: int | None = None, hidden_channels: int | None = None, out_channels: int | None = None, in_layers: int | None = None, out_layers: int | None = None, dropout_lin: float | List[float] = 0.0, dropout_conv: float = 0.0, act: str | Callable | None = 'relu', act_first: bool = False, act_kwargs: Dict[str, Any] | None = None, norm: str | Callable | None = 'batch_norm', norm_kwargs: Dict[str, Any] | None = None, plain_last: bool = False, bias: bool | List[bool] = True, **kwargs)

Bases: BaseNN

Decoupled structure for ACM conv.

Paper:

Revisiting Heterophily For Graph Neural Networks

Paper:

Complete the Missing Half: Augmenting Aggregation Filtering with Diversification for Graph Convolutional Networks

Ref:

https://github.com/SitaoLuan/ACM-GNN

Parameters:

• theta_scheme (str) – Channel list. “FBGNN”=”low-high”, “ACMGNN”=”low-high-id”, (“ACMGNN+”=”low-high-id-struct”, not implemented).

• weight_initializer (str, optional) – The initializer for the weight.

• conv (str) – args for BaseNN

• num_hops (int, default: 0) – args for BaseNN

• in_channels (int | None, default: None) – args for BaseNN

• hidden_channels (int | None, default: None) – args for BaseNN

• out_channels (int | None, default: None) – args for BaseNN

• in_layers (int | None, default: None) – args for BaseNN

• out_layers (int | None, default: None) – args for BaseNN

• dropout_lin (float | List[float], default: 0.0) – args for BaseNN

• dropout_conv (float, default: 0.0) – args for BaseNN

• lib_conv – args for BaseNN

• act (str | Callable | None, default: 'relu') – args for torch_geometric.nn.models.MLP.

• act_first (bool, default: False) – args for torch_geometric.nn.models.MLP.

• act_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• norm (str | Callable | None, default: 'batch_norm') – args for torch_geometric.nn.models.MLP.

• norm_kwargs (Dict[str, Any] | None, default: None) – args for torch_geometric.nn.models.MLP.

• plain_last (bool, default: False) – args for torch_geometric.nn.models.MLP.

• bias (bool | List[bool], default: True) – args for torch_geometric.nn.models.MLP.

• **kwargs – Additional arguments of pyg_spectral.nn.conv.

init_conv(conv: str, num_hops: int, lib: str, **kwargs) → MessagePassing

reset_parameters()

class CppCompFixed(in_layers: int | None = None, **kwargs)

Bases: PrecomputedFixed

Decoupled structure with C++ propagation precomputation. Fixed scalar propagation parameters and accumulating precompute results.

preprocess(x: Tensor, edge_index: Tensor | SparseTensor) → Tensor | SparseTensor

Preprocessing step that not counted in forward() overhead. Here mainly transforming graph adjacency to actual propagation matrix.

convolute(x: Tensor, edge_index: Tensor | SparseTensor) → Tensor

Decoupled propagation step for calling the convolutional module. Requires no variable transformation in conv.forward().

Returns:

embed (Tensor) – Precomputed node embeddings.