graph_bilinear_interdependence_layer

Bases: layer

A layer that combines graph interdependence with bilinear interdependence.

This layer integrates graph_bilinear_interdependence_head instances to simultaneously model graph-structured data relationships and bilinear dependencies.

Attributes:

Name	Type	Description
m	int	The input dimensionality of the layer.
n	int	The output dimensionality of the layer.
width	int	The number of graph_bilinear_interdependence_head instances in the layer.
name	str	The name of the layer.
channel_num	int	The number of channels in each graph_bilinear_interdependence_head.
graph	(graph, optional)	The graph structure to be used. If not provided, graph_file_path, nodes, and links must be provided.
graph_file_path	(str, optional)	Path to the file from which the graph can be loaded.
nodes	(list, optional)	A list of node identifiers in the graph.
links	(list, optional)	A list of edges in the graph.
directed	bool	Whether the graph is directed.
with_multihop	bool	Whether to enable multi-hop dependencies.
h	int	Number of hops for multi-hop dependencies.
accumulative	bool	Whether to accumulate dependencies across hops.
with_pagerank	bool	Whether to include PageRank-like dependencies in the layer.
c	float	Damping factor for PageRank dependencies.
require_data	bool	Whether data is required for the graph interdependence function.
require_parameters	bool	Whether parameters are required for the graph interdependence function.
normalization	bool	Whether to normalize the adjacency matrix.
normalization_mode	str	The normalization mode for the adjacency matrix ('row' or 'column').
self_dependence	bool	Whether to include self-dependencies in the graph structure.
with_dual_lphm_interdependence	bool	Whether to use dual LPHM bilinear interdependence.
with_lorr_interdependence	bool	Whether to use LORR bilinear interdependence.
r_interdependence	int	The rank for bilinear interdependence reconciliation.
with_dual_lphm	bool	Whether to use dual LPHM reconciliation.
with_lorr	bool	Whether to use LORR reconciliation.
r	int	The rank for parameter reconciliation.
with_residual	bool	Whether to include residual connections in the layer.
enable_bias	bool	Whether to enable bias terms in parameter reconciliation.
with_batch_norm	bool	Whether to apply batch normalization to the layer's output.
with_relu	bool	Whether to apply ReLU activation to the layer's output.
with_softmax	bool	Whether to apply softmax activation to the layer's output.
with_dropout	bool	Whether to apply dropout to the layer's output.
p	float	The dropout probability.
parameters_init_method	str	The initialization method for parameters.
device	str	The device on which to run the layer ('cpu' or 'cuda').

Methods:

Name	Description
__init__	Initializes the graph bilinear interdependence layer.

Source code in tinybig/layer/graph_based_layers.py

class graph_bilinear_interdependence_layer(layer):
    """
    A layer that combines graph interdependence with bilinear interdependence.

    This layer integrates `graph_bilinear_interdependence_head` instances to simultaneously model
    graph-structured data relationships and bilinear dependencies.

    Attributes
    ----------
    m : int
        The input dimensionality of the layer.
    n : int
        The output dimensionality of the layer.
    width : int
        The number of `graph_bilinear_interdependence_head` instances in the layer.
    name : str
        The name of the layer.
    channel_num : int
        The number of channels in each `graph_bilinear_interdependence_head`.
    graph : graph_class, optional
        The graph structure to be used. If not provided, `graph_file_path`, `nodes`, and `links` must be provided.
    graph_file_path : str, optional
        Path to the file from which the graph can be loaded.
    nodes : list, optional
        A list of node identifiers in the graph.
    links : list, optional
        A list of edges in the graph.
    directed : bool
        Whether the graph is directed.
    with_multihop : bool
        Whether to enable multi-hop dependencies.
    h : int
        Number of hops for multi-hop dependencies.
    accumulative : bool
        Whether to accumulate dependencies across hops.
    with_pagerank : bool
        Whether to include PageRank-like dependencies in the layer.
    c : float
        Damping factor for PageRank dependencies.
    require_data : bool
        Whether data is required for the graph interdependence function.
    require_parameters : bool
        Whether parameters are required for the graph interdependence function.
    normalization : bool
        Whether to normalize the adjacency matrix.
    normalization_mode : str
        The normalization mode for the adjacency matrix (`'row'` or `'column'`).
    self_dependence : bool
        Whether to include self-dependencies in the graph structure.
    with_dual_lphm_interdependence : bool
        Whether to use dual LPHM bilinear interdependence.
    with_lorr_interdependence : bool
        Whether to use LORR bilinear interdependence.
    r_interdependence : int
        The rank for bilinear interdependence reconciliation.
    with_dual_lphm : bool
        Whether to use dual LPHM reconciliation.
    with_lorr : bool
        Whether to use LORR reconciliation.
    r : int
        The rank for parameter reconciliation.
    with_residual : bool
        Whether to include residual connections in the layer.
    enable_bias : bool
        Whether to enable bias terms in parameter reconciliation.
    with_batch_norm : bool
        Whether to apply batch normalization to the layer's output.
    with_relu : bool
        Whether to apply ReLU activation to the layer's output.
    with_softmax : bool
        Whether to apply softmax activation to the layer's output.
    with_dropout : bool
        Whether to apply dropout to the layer's output.
    p : float
        The dropout probability.
    parameters_init_method : str
        The initialization method for parameters.
    device : str
        The device on which to run the layer (`'cpu'` or `'cuda'`).

    Methods
    -------
    __init__(...)
        Initializes the graph bilinear interdependence layer.
    """
    def __init__(
        self,
        m: int, n: int,
        width: int = 1,
        name: str = 'graph_bilinear_interdependence_layer',
        channel_num: int = 1,
        # graph structure parameters
        graph: graph_class = None,
        graph_file_path: str = None,
        nodes: list = None,
        links: list = None,
        directed: bool = False,
        # graph interdependence function parameters
        with_multihop: bool = False, h: int = 1, accumulative: bool = False,
        with_pagerank: bool = False, c: float = 0.15,
        require_data: bool = False,
        require_parameters: bool = False,
        # adj matrix processing parameters
        normalization: bool = False,
        normalization_mode: str = 'column',
        self_dependence: bool = True,
        # bilinear interdependence function parameters
        with_dual_lphm_interdependence: bool = False,
        with_lorr_interdependence: bool = False, r_interdependence: int = 3,
        # parameter reconciliation and remainder functions
        with_dual_lphm: bool = False,
        with_lorr: bool = False, r: int = 3,
        with_residual: bool = False,
        enable_bias: bool = False,
        # output processing parameters
        with_batch_norm: bool = False,
        with_relu: bool = True,
        with_softmax: bool = True,
        with_dropout: bool = True, p: float = 0.5,
        # other parameters
        parameters_init_method: str = 'xavier_normal',
        device: str = 'cpu', *args, **kwargs
    ):
        """
        Initialize the Graph Bilinear Interdependence Layer.

        This method creates a graph bilinear interdependence layer that combines graph-based
        dependencies with bilinear interdependence functions. The layer consists of multiple
        `graph_bilinear_interdependence_head` instances, which are configured based on the
        provided parameters.

        Parameters
        ----------
        m : int
            The input dimensionality of the layer.
        n : int
            The output dimensionality of the layer.
        width : int, optional
            The number of `graph_bilinear_interdependence_head` instances in the layer. Default is 1.
        name : str, optional
            The name of the layer. Default is `'graph_bilinear_interdependence_layer'`.
        channel_num : int, optional
            The number of channels in each head. Default is 1.
        graph : graph_class, optional
            A pre-loaded graph structure. If provided, this takes precedence over `graph_file_path`, `nodes`, and `links`.
        graph_file_path : str, optional
            Path to the file containing the graph structure.
        nodes : list, optional
            A list of node identifiers.
        links : list, optional
            A list of edges representing connections between nodes.
        directed : bool, optional
            Indicates whether the graph is directed. Default is `False`.
        with_multihop : bool, optional
            If `True`, enables multi-hop dependencies in the graph interdependence function. Default is `False`.
        h : int, optional
            The number of hops to consider for multi-hop dependencies. Default is 1.
        accumulative : bool, optional
            If `True`, accumulates dependencies across multiple hops. Default is `False`.
        with_pagerank : bool, optional
            If `True`, includes PageRank-like dependencies in the graph interdependence function. Default is `False`.
        c : float, optional
            Damping factor for PageRank dependencies. Default is 0.15.
        require_data : bool, optional
            If `True`, requires data for the graph interdependence function. Default is `False`.
        require_parameters : bool, optional
            If `True`, requires parameters for the graph interdependence function. Default is `False`.
        normalization : bool, optional
            If `True`, normalizes the adjacency matrix of the graph. Default is `False`.
        normalization_mode : str, optional
            Specifies the normalization mode (`'row'` or `'column'`). Default is `'column'`.
        self_dependence : bool, optional
            If `True`, includes self-loops (self-dependencies) in the graph structure. Default is `True`.
        with_dual_lphm_interdependence : bool, optional
            If `True`, uses dual LPHM bilinear interdependence. Default is `False`.
        with_lorr_interdependence : bool, optional
            If `True`, uses LORR bilinear interdependence. Default is `False`.
        r_interdependence : int, optional
            The rank for bilinear interdependence parameterization. Default is 3.
        with_dual_lphm : bool, optional
            If `True`, applies dual LPHM reconciliation for parameter fabrication. Default is `False`.
        with_lorr : bool, optional
            If `True`, applies LORR reconciliation for parameter fabrication. Default is `False`.
        r : int, optional
            The rank used for parameter reconciliation. Default is 3.
        with_residual : bool, optional
            If `True`, includes residual connections. Default is `False`.
        enable_bias : bool, optional
            If `True`, enables bias terms during parameter reconciliation. Default is `False`.
        with_batch_norm : bool, optional
            If `True`, applies batch normalization to the output. Default is `False`.
        with_relu : bool, optional
            If `True`, applies a ReLU activation to the output. Default is `True`.
        with_softmax : bool, optional
            If `True`, applies a softmax activation to the output. Default is `True`.
        with_dropout : bool, optional
            If `True`, applies dropout to the output. Default is `True`.
        p : float, optional
            The dropout probability. Default is 0.5.
        parameters_init_method : str, optional
            Specifies the initialization method for parameters. Default is `'xavier_normal'`.
        device : str, optional
            The device on which to run the layer (`'cpu'` or `'cuda'`). Default is `'cpu'`.

        Raises
        ------
        ValueError
            If a graph structure is not provided via `graph`, `graph_file_path`, `nodes`, or `links`.

        Notes
        -----
        - This layer supports both graph-based interdependence and bilinear interdependence,
          combining the two through a fusion mechanism.
        - If both `graph` and `graph_file_path` are provided, `graph` takes precedence.
        """
        print('* graph_bilinear_interdependence_layer, width:', width)
        heads = [
            graph_bilinear_interdependence_head(
                m=m, n=n,
                channel_num=channel_num,
                # -------------------
                graph=graph,
                graph_file_path=graph_file_path,
                nodes=nodes,
                links=links,
                directed=directed,
                # -------------------
                with_multihop=with_multihop, h=h, accumulative=accumulative,
                with_pagerank=with_pagerank, c=c,
                require_data=require_data,
                require_parameters=require_parameters,
                # -------------------
                normalization=normalization,
                normalization_mode=normalization_mode,
                self_dependence=self_dependence,
                # -------------------
                with_dual_lphm_interdependence=with_dual_lphm_interdependence,
                with_lorr_interdependence=with_lorr_interdependence, r_interdependence=r_interdependence,
                # -------------------
                with_dual_lphm=with_dual_lphm,
                with_lorr=with_lorr, r=r,
                with_residual=with_residual,
                enable_bias=enable_bias,
                # -------------------
                with_batch_norm=with_batch_norm,
                with_relu=with_relu,
                with_softmax=with_softmax,
                with_dropout=with_dropout, p=p,
                # -------------------
                parameters_init_method=parameters_init_method,
                device=device, *args, **kwargs
            )
        ] * width
        print('--------------------------')
        super().__init__(name=name, m=m, n=n, heads=heads, device=device, *args, **kwargs)

__init__(m, n, width=1, name='graph_bilinear_interdependence_layer', channel_num=1, graph=None, graph_file_path=None, nodes=None, links=None, directed=False, with_multihop=False, h=1, accumulative=False, with_pagerank=False, c=0.15, require_data=False, require_parameters=False, normalization=False, normalization_mode='column', self_dependence=True, with_dual_lphm_interdependence=False, with_lorr_interdependence=False, r_interdependence=3, with_dual_lphm=False, with_lorr=False, r=3, with_residual=False, enable_bias=False, with_batch_norm=False, with_relu=True, with_softmax=True, with_dropout=True, p=0.5, parameters_init_method='xavier_normal', device='cpu', *args, **kwargs)

Initialize the Graph Bilinear Interdependence Layer.

This method creates a graph bilinear interdependence layer that combines graph-based dependencies with bilinear interdependence functions. The layer consists of multiple graph_bilinear_interdependence_head instances, which are configured based on the provided parameters.

Parameters:

Name	Type	Description	Default
m	int	The input dimensionality of the layer.	required
n	int	The output dimensionality of the layer.	required
width	int	The number of graph_bilinear_interdependence_head instances in the layer. Default is 1.	1
name	str	The name of the layer. Default is 'graph_bilinear_interdependence_layer'.	'graph_bilinear_interdependence_layer'
channel_num	int	The number of channels in each head. Default is 1.	1
graph	graph	A pre-loaded graph structure. If provided, this takes precedence over graph_file_path, nodes, and links.	None
graph_file_path	str	Path to the file containing the graph structure.	None
nodes	list	A list of node identifiers.	None
links	list	A list of edges representing connections between nodes.	None
directed	bool	Indicates whether the graph is directed. Default is False.	False
with_multihop	bool	If True, enables multi-hop dependencies in the graph interdependence function. Default is False.	False
h	int	The number of hops to consider for multi-hop dependencies. Default is 1.	1
accumulative	bool	If True, accumulates dependencies across multiple hops. Default is False.	False
with_pagerank	bool	If True, includes PageRank-like dependencies in the graph interdependence function. Default is False.	False
c	float	Damping factor for PageRank dependencies. Default is 0.15.	0.15
require_data	bool	If True, requires data for the graph interdependence function. Default is False.	False
require_parameters	bool	If True, requires parameters for the graph interdependence function. Default is False.	False
normalization	bool	If True, normalizes the adjacency matrix of the graph. Default is False.	False
normalization_mode	str	Specifies the normalization mode ('row' or 'column'). Default is 'column'.	'column'
self_dependence	bool	If True, includes self-loops (self-dependencies) in the graph structure. Default is True.	True
with_dual_lphm_interdependence	bool	If True, uses dual LPHM bilinear interdependence. Default is False.	False
with_lorr_interdependence	bool	If True, uses LORR bilinear interdependence. Default is False.	False
r_interdependence	int	The rank for bilinear interdependence parameterization. Default is 3.	3
with_dual_lphm	bool	If True, applies dual LPHM reconciliation for parameter fabrication. Default is False.	False
with_lorr	bool	If True, applies LORR reconciliation for parameter fabrication. Default is False.	False
r	int	The rank used for parameter reconciliation. Default is 3.	3
with_residual	bool	If True, includes residual connections. Default is False.	False
enable_bias	bool	If True, enables bias terms during parameter reconciliation. Default is False.	False
with_batch_norm	bool	If True, applies batch normalization to the output. Default is False.	False
with_relu	bool	If True, applies a ReLU activation to the output. Default is True.	True
with_softmax	bool	If True, applies a softmax activation to the output. Default is True.	True
with_dropout	bool	If True, applies dropout to the output. Default is True.	True
p	float	The dropout probability. Default is 0.5.	0.5
parameters_init_method	str	Specifies the initialization method for parameters. Default is 'xavier_normal'.	'xavier_normal'
device	str	The device on which to run the layer ('cpu' or 'cuda'). Default is 'cpu'.	'cpu'

Raises:

Type	Description
ValueError	If a graph structure is not provided via graph, graph_file_path, nodes, or links.

Notes

• This layer supports both graph-based interdependence and bilinear interdependence, combining the two through a fusion mechanism.
• If both graph and graph_file_path are provided, graph takes precedence.

Source code in tinybig/layer/graph_based_layers.py

def __init__(
    self,
    m: int, n: int,
    width: int = 1,
    name: str = 'graph_bilinear_interdependence_layer',
    channel_num: int = 1,
    # graph structure parameters
    graph: graph_class = None,
    graph_file_path: str = None,
    nodes: list = None,
    links: list = None,
    directed: bool = False,
    # graph interdependence function parameters
    with_multihop: bool = False, h: int = 1, accumulative: bool = False,
    with_pagerank: bool = False, c: float = 0.15,
    require_data: bool = False,
    require_parameters: bool = False,
    # adj matrix processing parameters
    normalization: bool = False,
    normalization_mode: str = 'column',
    self_dependence: bool = True,
    # bilinear interdependence function parameters
    with_dual_lphm_interdependence: bool = False,
    with_lorr_interdependence: bool = False, r_interdependence: int = 3,
    # parameter reconciliation and remainder functions
    with_dual_lphm: bool = False,
    with_lorr: bool = False, r: int = 3,
    with_residual: bool = False,
    enable_bias: bool = False,
    # output processing parameters
    with_batch_norm: bool = False,
    with_relu: bool = True,
    with_softmax: bool = True,
    with_dropout: bool = True, p: float = 0.5,
    # other parameters
    parameters_init_method: str = 'xavier_normal',
    device: str = 'cpu', *args, **kwargs
):
    """
    Initialize the Graph Bilinear Interdependence Layer.

    This method creates a graph bilinear interdependence layer that combines graph-based
    dependencies with bilinear interdependence functions. The layer consists of multiple
    `graph_bilinear_interdependence_head` instances, which are configured based on the
    provided parameters.

    Parameters
    ----------
    m : int
        The input dimensionality of the layer.
    n : int
        The output dimensionality of the layer.
    width : int, optional
        The number of `graph_bilinear_interdependence_head` instances in the layer. Default is 1.
    name : str, optional
        The name of the layer. Default is `'graph_bilinear_interdependence_layer'`.
    channel_num : int, optional
        The number of channels in each head. Default is 1.
    graph : graph_class, optional
        A pre-loaded graph structure. If provided, this takes precedence over `graph_file_path`, `nodes`, and `links`.
    graph_file_path : str, optional
        Path to the file containing the graph structure.
    nodes : list, optional
        A list of node identifiers.
    links : list, optional
        A list of edges representing connections between nodes.
    directed : bool, optional
        Indicates whether the graph is directed. Default is `False`.
    with_multihop : bool, optional
        If `True`, enables multi-hop dependencies in the graph interdependence function. Default is `False`.
    h : int, optional
        The number of hops to consider for multi-hop dependencies. Default is 1.
    accumulative : bool, optional
        If `True`, accumulates dependencies across multiple hops. Default is `False`.
    with_pagerank : bool, optional
        If `True`, includes PageRank-like dependencies in the graph interdependence function. Default is `False`.
    c : float, optional
        Damping factor for PageRank dependencies. Default is 0.15.
    require_data : bool, optional
        If `True`, requires data for the graph interdependence function. Default is `False`.
    require_parameters : bool, optional
        If `True`, requires parameters for the graph interdependence function. Default is `False`.
    normalization : bool, optional
        If `True`, normalizes the adjacency matrix of the graph. Default is `False`.
    normalization_mode : str, optional
        Specifies the normalization mode (`'row'` or `'column'`). Default is `'column'`.
    self_dependence : bool, optional
        If `True`, includes self-loops (self-dependencies) in the graph structure. Default is `True`.
    with_dual_lphm_interdependence : bool, optional
        If `True`, uses dual LPHM bilinear interdependence. Default is `False`.
    with_lorr_interdependence : bool, optional
        If `True`, uses LORR bilinear interdependence. Default is `False`.
    r_interdependence : int, optional
        The rank for bilinear interdependence parameterization. Default is 3.
    with_dual_lphm : bool, optional
        If `True`, applies dual LPHM reconciliation for parameter fabrication. Default is `False`.
    with_lorr : bool, optional
        If `True`, applies LORR reconciliation for parameter fabrication. Default is `False`.
    r : int, optional
        The rank used for parameter reconciliation. Default is 3.
    with_residual : bool, optional
        If `True`, includes residual connections. Default is `False`.
    enable_bias : bool, optional
        If `True`, enables bias terms during parameter reconciliation. Default is `False`.
    with_batch_norm : bool, optional
        If `True`, applies batch normalization to the output. Default is `False`.
    with_relu : bool, optional
        If `True`, applies a ReLU activation to the output. Default is `True`.
    with_softmax : bool, optional
        If `True`, applies a softmax activation to the output. Default is `True`.
    with_dropout : bool, optional
        If `True`, applies dropout to the output. Default is `True`.
    p : float, optional
        The dropout probability. Default is 0.5.
    parameters_init_method : str, optional
        Specifies the initialization method for parameters. Default is `'xavier_normal'`.
    device : str, optional
        The device on which to run the layer (`'cpu'` or `'cuda'`). Default is `'cpu'`.

    Raises
    ------
    ValueError
        If a graph structure is not provided via `graph`, `graph_file_path`, `nodes`, or `links`.

    Notes
    -----
    - This layer supports both graph-based interdependence and bilinear interdependence,
      combining the two through a fusion mechanism.
    - If both `graph` and `graph_file_path` are provided, `graph` takes precedence.
    """
    print('* graph_bilinear_interdependence_layer, width:', width)
    heads = [
        graph_bilinear_interdependence_head(
            m=m, n=n,
            channel_num=channel_num,
            # -------------------
            graph=graph,
            graph_file_path=graph_file_path,
            nodes=nodes,
            links=links,
            directed=directed,
            # -------------------
            with_multihop=with_multihop, h=h, accumulative=accumulative,
            with_pagerank=with_pagerank, c=c,
            require_data=require_data,
            require_parameters=require_parameters,
            # -------------------
            normalization=normalization,
            normalization_mode=normalization_mode,
            self_dependence=self_dependence,
            # -------------------
            with_dual_lphm_interdependence=with_dual_lphm_interdependence,
            with_lorr_interdependence=with_lorr_interdependence, r_interdependence=r_interdependence,
            # -------------------
            with_dual_lphm=with_dual_lphm,
            with_lorr=with_lorr, r=r,
            with_residual=with_residual,
            enable_bias=enable_bias,
            # -------------------
            with_batch_norm=with_batch_norm,
            with_relu=with_relu,
            with_softmax=with_softmax,
            with_dropout=with_dropout, p=p,
            # -------------------
            parameters_init_method=parameters_init_method,
            device=device, *args, **kwargs
        )
    ] * width
    print('--------------------------')
    super().__init__(name=name, m=m, n=n, heads=heads, device=device, *args, **kwargs)