nip.graph_isomorphism.environment.AdjacencyMatrixSpec#

class nip.graph_isomorphism.environment.AdjacencyMatrixSpec(max_num_nodes: int, shape: Size | None = None, device: device | str | int | None = None, dtype: str | dtype = torch.int32)[source]#

A specification of the space of adjacency matrices.

This represents a space of adjacency matrices with a fixed number of nodes, for use in an RL environment.

Parameters:

• max_num_nodes (int) – The maximum number of nodes in the graph.

• shape (torch.Size, optional) – The shape of the adjacency matrix. If None, the shape will be (max_num_nodes, max_num_nodes).

• device (torch.device, optional) – The device on which the adjacency matrix spec should be stored.

• dtype (torch.dtype, optional) – The dtype of the adjacency matrix spec.

Methods Summary

__init__(max_num_nodes[, shape, device, dtype])
_project(val)	Project a value to the space of valid adjacency matrices.
clone()	Create a copy of the adjacency matrix spec.
is_in(val)	Check if a value is a valid adjacency matrix.
rand([shape])	Generate a random 1/2 Erdos-Renyi adjacency matrix.
to(dest)	Move the adjacency matrix spec to a new device or dtype.

Attributes

SPEC_HANDLED_FUNCTIONS
device
domain
dtype
ndim
shape
space

Methods

__init__(max_num_nodes: int, shape: Size | None = None, device: device | str | int | None = None, dtype: str | dtype = torch.int32)[source]#

_project(val: Tensor) → Tensor[source]#

Project a value to the space of valid adjacency matrices.

Symmetrizes the matrix, sets the diagonal to zero and rounds the values to 0 or 1.

Parameters:

val (torch.Tensor) – The value to project.

Returns:

projected_val (torch.Tensor) – The projected value.

clone() → AdjacencyMatrixSpec[source]#

Create a copy of the adjacency matrix spec.

is_in(val: Tensor) → bool[source]#

Check if a value is a valid adjacency matrix.

Parameters:

val (torch.Tensor) – The value to check.

Returns:

is_in (bool) – Whether the value is a valid adjacency matrix.

rand(shape: list[int] | Size | None = None) → Tensor[source]#

Generate a random 1/2 Erdos-Renyi adjacency matrix.

Parameters:

shape (list[int] | torch.Size, optional) – The batch shape of the adjacency matrix. If None, the batch shape will be ().

Returns:

adjacency (torch.Tensor) – A random adjacency matrix.

to(dest: dtype | device | str | int) → TensorSpec[source]#

Move the adjacency matrix spec to a new device or dtype.

Parameters:

dest (torch.dtype | torch.device | str | int) – The destination device or dtype.

Returns:

adjacency_matrix_spec (AdjacencyMatrixSpec) – The adjacency matrix spec on the new device or dtype.