import os.path as osp
import warnings
from abc import abstractmethod
from inspect import Parameter
from typing import (
Any,
Callable,
Dict,
Final,
List,
Optional,
OrderedDict,
Set,
Tuple,
Union,
)
import torch
from torch import Tensor
from torch.utils.hooks import RemovableHandle
from torch_geometric import EdgeIndex, is_compiling
from torch_geometric.index import ptr2index
from torch_geometric.inspector import Inspector, Signature
from torch_geometric.nn.aggr import Aggregation
from torch_geometric.nn.resolver import aggregation_resolver as aggr_resolver
from torch_geometric.template import module_from_template
from torch_geometric.typing import Adj, Size, SparseTensor
from torch_geometric.utils import (
is_sparse,
is_torch_sparse_tensor,
to_edge_index,
)
FUSE_AGGRS = {'add', 'sum', 'mean', 'min', 'max'}
HookDict = OrderedDict[int, Callable]
[docs]class MessagePassing(torch.nn.Module):
r"""Base class for creating message passing layers.
Message passing layers follow the form
.. math::
\mathbf{x}_i^{\prime} = \gamma_{\mathbf{\Theta}} \left( \mathbf{x}_i,
\bigoplus_{j \in \mathcal{N}(i)} \, \phi_{\mathbf{\Theta}}
\left(\mathbf{x}_i, \mathbf{x}_j,\mathbf{e}_{j,i}\right) \right),
where :math:`\bigoplus` denotes a differentiable, permutation invariant
function, *e.g.*, sum, mean, min, max or mul, and
:math:`\gamma_{\mathbf{\Theta}}` and :math:`\phi_{\mathbf{\Theta}}` denote
differentiable functions such as MLPs.
See `here <https://pytorch-geometric.readthedocs.io/en/latest/tutorial/
create_gnn.html>`__ for the accompanying tutorial.
Args:
aggr (str or [str] or Aggregation, optional): The aggregation scheme
to use, *e.g.*, :obj:`"sum"` :obj:`"mean"`, :obj:`"min"`,
:obj:`"max"` or :obj:`"mul"`.
In addition, can be any
:class:`~torch_geometric.nn.aggr.Aggregation` module (or any string
that automatically resolves to it).
If given as a list, will make use of multiple aggregations in which
different outputs will get concatenated in the last dimension.
If set to :obj:`None`, the :class:`MessagePassing` instantiation is
expected to implement its own aggregation logic via
:meth:`aggregate`. (default: :obj:`"add"`)
aggr_kwargs (Dict[str, Any], optional): Arguments passed to the
respective aggregation function in case it gets automatically
resolved. (default: :obj:`None`)
flow (str, optional): The flow direction of message passing
(:obj:`"source_to_target"` or :obj:`"target_to_source"`).
(default: :obj:`"source_to_target"`)
node_dim (int, optional): The axis along which to propagate.
(default: :obj:`-2`)
decomposed_layers (int, optional): The number of feature decomposition
layers, as introduced in the `"Optimizing Memory Efficiency of
Graph Neural Networks on Edge Computing Platforms"
<https://arxiv.org/abs/2104.03058>`_ paper.
Feature decomposition reduces the peak memory usage by slicing
the feature dimensions into separated feature decomposition layers
during GNN aggregation.
This method can accelerate GNN execution on CPU-based platforms
(*e.g.*, 2-3x speedup on the
:class:`~torch_geometric.datasets.Reddit` dataset) for common GNN
models such as :class:`~torch_geometric.nn.models.GCN`,
:class:`~torch_geometric.nn.models.GraphSAGE`,
:class:`~torch_geometric.nn.models.GIN`, etc.
However, this method is not applicable to all GNN operators
available, in particular for operators in which message computation
can not easily be decomposed, *e.g.* in attention-based GNNs.
The selection of the optimal value of :obj:`decomposed_layers`
depends both on the specific graph dataset and available hardware
resources.
A value of :obj:`2` is suitable in most cases.
Although the peak memory usage is directly associated with the
granularity of feature decomposition, the same is not necessarily
true for execution speedups. (default: :obj:`1`)
"""
special_args: Set[str] = {
'edge_index', 'adj_t', 'edge_index_i', 'edge_index_j', 'size',
'size_i', 'size_j', 'ptr', 'index', 'dim_size'
}
# Supports `message_and_aggregate` via `EdgeIndex`.
# TODO Remove once migration is finished.
SUPPORTS_FUSED_EDGE_INDEX: Final[bool] = False
def __init__(
self,
aggr: Optional[Union[str, List[str], Aggregation]] = 'sum',
*,
aggr_kwargs: Optional[Dict[str, Any]] = None,
flow: str = "source_to_target",
node_dim: int = -2,
decomposed_layers: int = 1,
) -> None:
super().__init__()
if flow not in ['source_to_target', 'target_to_source']:
raise ValueError(f"Expected 'flow' to be either 'source_to_target'"
f" or 'target_to_source' (got '{flow}')")
# Cast `aggr` into a string representation for backward compatibility:
self.aggr: Optional[Union[str, List[str]]]
if aggr is None:
self.aggr = None
elif isinstance(aggr, (str, Aggregation)):
self.aggr = str(aggr)
elif isinstance(aggr, (tuple, list)):
self.aggr = [str(x) for x in aggr]
self.aggr_module = aggr_resolver(aggr, **(aggr_kwargs or {}))
self.flow = flow
self.node_dim = node_dim
# Collect attribute names requested in message passing hooks:
self.inspector = Inspector(self.__class__)
self.inspector.inspect_signature(self.message)
self.inspector.inspect_signature(self.aggregate, exclude=[0, 'aggr'])
self.inspector.inspect_signature(self.message_and_aggregate, [0])
self.inspector.inspect_signature(self.update, exclude=[0])
self.inspector.inspect_signature(self.edge_update)
self._user_args: List[str] = self.inspector.get_flat_param_names(
['message', 'aggregate', 'update'], exclude=self.special_args)
self._fused_user_args: List[str] = self.inspector.get_flat_param_names(
['message_and_aggregate', 'update'], exclude=self.special_args)
self._edge_user_args: List[str] = self.inspector.get_param_names(
'edge_update', exclude=self.special_args)
# Support for "fused" message passing:
self.fuse = self.inspector.implements('message_and_aggregate')
if self.aggr is not None:
self.fuse &= isinstance(self.aggr, str) and self.aggr in FUSE_AGGRS
# Hooks:
self._propagate_forward_pre_hooks: HookDict = OrderedDict()
self._propagate_forward_hooks: HookDict = OrderedDict()
self._message_forward_pre_hooks: HookDict = OrderedDict()
self._message_forward_hooks: HookDict = OrderedDict()
self._aggregate_forward_pre_hooks: HookDict = OrderedDict()
self._aggregate_forward_hooks: HookDict = OrderedDict()
self._message_and_aggregate_forward_pre_hooks: HookDict = OrderedDict()
self._message_and_aggregate_forward_hooks: HookDict = OrderedDict()
self._edge_update_forward_pre_hooks: HookDict = OrderedDict()
self._edge_update_forward_hooks: HookDict = OrderedDict()
# Set jittable `propagate` and `edge_updater` function templates:
self._set_jittable_templates()
# Explainability:
self._explain: Optional[bool] = None
self._edge_mask: Optional[Tensor] = None
self._loop_mask: Optional[Tensor] = None
self._apply_sigmoid: bool = True
# Inference Decomposition:
self._decomposed_layers = 1
self.decomposed_layers = decomposed_layers
[docs] def reset_parameters(self) -> None:
r"""Resets all learnable parameters of the module."""
if self.aggr_module is not None:
self.aggr_module.reset_parameters()
def __setstate__(self, data: Dict[str, Any]) -> None:
self.inspector = data['inspector']
self.fuse = data['fuse']
self._set_jittable_templates()
super().__setstate__(data)
def __repr__(self) -> str:
channels_repr = ''
if hasattr(self, 'in_channels') and hasattr(self, 'out_channels'):
channels_repr = f'{self.in_channels}, {self.out_channels}'
elif hasattr(self, 'channels'):
channels_repr = f'{self.channels}'
return f'{self.__class__.__name__}({channels_repr})'
# Utilities ###############################################################
def _check_input(
self,
edge_index: Union[Tensor, SparseTensor],
size: Optional[Tuple[Optional[int], Optional[int]]],
) -> List[Optional[int]]:
if not torch.jit.is_scripting() and isinstance(edge_index, EdgeIndex):
return [edge_index.num_rows, edge_index.num_cols]
if is_sparse(edge_index):
if self.flow == 'target_to_source':
raise ValueError(
'Flow direction "target_to_source" is invalid for '
'message propagation via `torch_sparse.SparseTensor` '
'or `torch.sparse.Tensor`. If you really want to make '
'use of a reverse message passing flow, pass in the '
'transposed sparse tensor to the message passing module, '
'e.g., `adj_t.t()`.')
if isinstance(edge_index, SparseTensor):
return [edge_index.size(1), edge_index.size(0)]
return [edge_index.size(1), edge_index.size(0)]
elif isinstance(edge_index, Tensor):
int_dtypes = (torch.uint8, torch.int8, torch.int16, torch.int32,
torch.int64)
if edge_index.dtype not in int_dtypes:
raise ValueError(f"Expected 'edge_index' to be of integer "
f"type (got '{edge_index.dtype}')")
if edge_index.dim() != 2:
raise ValueError(f"Expected 'edge_index' to be two-dimensional"
f" (got {edge_index.dim()} dimensions)")
if not torch.jit.is_tracing() and edge_index.size(0) != 2:
raise ValueError(f"Expected 'edge_index' to have size '2' in "
f"the first dimension (got "
f"'{edge_index.size(0)}')")
return list(size) if size is not None else [None, None]
raise ValueError(
'`MessagePassing.propagate` only supports integer tensors of '
'shape `[2, num_messages]`, `torch_sparse.SparseTensor` or '
'`torch.sparse.Tensor` for argument `edge_index`.')
def _set_size(
self,
size: List[Optional[int]],
dim: int,
src: Tensor,
) -> None:
the_size = size[dim]
if the_size is None:
size[dim] = src.size(self.node_dim)
elif the_size != src.size(self.node_dim):
raise ValueError(
f'Encountered tensor with size {src.size(self.node_dim)} in '
f'dimension {self.node_dim}, but expected size {the_size}.')
def _index_select(self, src: Tensor, index) -> Tensor:
if torch.jit.is_scripting() or is_compiling():
return src.index_select(self.node_dim, index)
else:
return self._index_select_safe(src, index)
def _index_select_safe(self, src: Tensor, index: Tensor) -> Tensor:
try:
return src.index_select(self.node_dim, index)
except (IndexError, RuntimeError) as e:
if index.numel() > 0 and index.min() < 0:
raise IndexError(
f"Found negative indices in 'edge_index' (got "
f"{index.min().item()}). Please ensure that all "
f"indices in 'edge_index' point to valid indices "
f"in the interval [0, {src.size(self.node_dim)}) in "
f"your node feature matrix and try again.")
if (index.numel() > 0 and index.max() >= src.size(self.node_dim)):
raise IndexError(
f"Found indices in 'edge_index' that are larger "
f"than {src.size(self.node_dim) - 1} (got "
f"{index.max().item()}). Please ensure that all "
f"indices in 'edge_index' point to valid indices "
f"in the interval [0, {src.size(self.node_dim)}) in "
f"your node feature matrix and try again.")
raise e
def _lift(
self,
src: Tensor,
edge_index: Union[Tensor, SparseTensor],
dim: int,
) -> Tensor:
if not torch.jit.is_scripting() and is_torch_sparse_tensor(edge_index):
assert dim == 0 or dim == 1
if edge_index.layout == torch.sparse_coo:
index = edge_index._indices()[1 - dim]
elif edge_index.layout == torch.sparse_csr:
if dim == 0:
index = edge_index.col_indices()
else:
index = ptr2index(edge_index.crow_indices())
elif edge_index.layout == torch.sparse_csc:
if dim == 0:
index = ptr2index(edge_index.ccol_indices())
else:
index = edge_index.row_indices()
else:
raise ValueError(f"Unsupported sparse tensor layout "
f"(got '{edge_index.layout}')")
return src.index_select(self.node_dim, index)
elif isinstance(edge_index, Tensor):
if torch.jit.is_scripting(): # Try/catch blocks are not supported.
index = edge_index[dim]
return src.index_select(self.node_dim, index)
return self._index_select(src, edge_index[dim])
elif isinstance(edge_index, SparseTensor):
row, col, _ = edge_index.coo()
if dim == 0:
return src.index_select(self.node_dim, col)
elif dim == 1:
return src.index_select(self.node_dim, row)
raise ValueError(
'`MessagePassing.propagate` only supports integer tensors of '
'shape `[2, num_messages]`, `torch_sparse.SparseTensor` '
'or `torch.sparse.Tensor` for argument `edge_index`.')
def _collect(
self,
args: Set[str],
edge_index: Union[Tensor, SparseTensor],
size: List[Optional[int]],
kwargs: Dict[str, Any],
) -> Dict[str, Any]:
i, j = (1, 0) if self.flow == 'source_to_target' else (0, 1)
out = {}
for arg in args:
if arg[-2:] not in ['_i', '_j']:
out[arg] = kwargs.get(arg, Parameter.empty)
else:
dim = j if arg[-2:] == '_j' else i
data = kwargs.get(arg[:-2], Parameter.empty)
if isinstance(data, (tuple, list)):
assert len(data) == 2
if isinstance(data[1 - dim], Tensor):
self._set_size(size, 1 - dim, data[1 - dim])
data = data[dim]
if isinstance(data, Tensor):
self._set_size(size, dim, data)
data = self._lift(data, edge_index, dim)
out[arg] = data
if is_torch_sparse_tensor(edge_index):
indices, values = to_edge_index(edge_index)
out['adj_t'] = edge_index
out['edge_index'] = None
out['edge_index_i'] = indices[0]
out['edge_index_j'] = indices[1]
out['ptr'] = None # TODO Get `rowptr` from CSR representation.
if out.get('edge_weight', None) is None:
out['edge_weight'] = values
if out.get('edge_attr', None) is None:
out['edge_attr'] = None if values.dim() == 1 else values
if out.get('edge_type', None) is None:
out['edge_type'] = values
elif isinstance(edge_index, Tensor):
out['adj_t'] = None
out['edge_index'] = edge_index
out['edge_index_i'] = edge_index[i]
out['edge_index_j'] = edge_index[j]
out['ptr'] = None
if isinstance(edge_index, EdgeIndex):
if i == 0 and edge_index.is_sorted_by_row:
(out['ptr'], _), _ = edge_index.get_csr()
elif i == 1 and edge_index.is_sorted_by_col:
(out['ptr'], _), _ = edge_index.get_csc()
elif isinstance(edge_index, SparseTensor):
row, col, value = edge_index.coo()
rowptr, _, _ = edge_index.csr()
out['adj_t'] = edge_index
out['edge_index'] = None
out['edge_index_i'] = row
out['edge_index_j'] = col
out['ptr'] = rowptr
if out.get('edge_weight', None) is None:
out['edge_weight'] = value
if out.get('edge_attr', None) is None:
out['edge_attr'] = value
if out.get('edge_type', None) is None:
out['edge_type'] = value
out['index'] = out['edge_index_i']
out['size'] = size
out['size_i'] = size[i] if size[i] is not None else size[j]
out['size_j'] = size[j] if size[j] is not None else size[i]
out['dim_size'] = out['size_i']
return out
# Message Passing #########################################################
[docs] def forward(self, *args: Any, **kwargs: Any) -> Any:
r"""Runs the forward pass of the module."""
[docs] def propagate(
self,
edge_index: Adj,
size: Size = None,
**kwargs: Any,
) -> Tensor:
r"""The initial call to start propagating messages.
Args:
edge_index (torch.Tensor or SparseTensor): A :class:`torch.Tensor`,
a :class:`torch_sparse.SparseTensor` or a
:class:`torch.sparse.Tensor` that defines the underlying
graph connectivity/message passing flow.
:obj:`edge_index` holds the indices of a general (sparse)
assignment matrix of shape :obj:`[N, M]`.
If :obj:`edge_index` is a :obj:`torch.Tensor`, its :obj:`dtype`
should be :obj:`torch.long` and its shape needs to be defined
as :obj:`[2, num_messages]` where messages from nodes in
:obj:`edge_index[0]` are sent to nodes in :obj:`edge_index[1]`
(in case :obj:`flow="source_to_target"`).
If :obj:`edge_index` is a :class:`torch_sparse.SparseTensor` or
a :class:`torch.sparse.Tensor`, its sparse indices
:obj:`(row, col)` should relate to :obj:`row = edge_index[1]`
and :obj:`col = edge_index[0]`.
The major difference between both formats is that we need to
input the *transposed* sparse adjacency matrix into
:meth:`propagate`.
size ((int, int), optional): The size :obj:`(N, M)` of the
assignment matrix in case :obj:`edge_index` is a
:class:`torch.Tensor`.
If set to :obj:`None`, the size will be automatically inferred
and assumed to be quadratic.
This argument is ignored in case :obj:`edge_index` is a
:class:`torch_sparse.SparseTensor` or
a :class:`torch.sparse.Tensor`. (default: :obj:`None`)
**kwargs: Any additional data which is needed to construct and
aggregate messages, and to update node embeddings.
"""
decomposed_layers = 1 if self.explain else self.decomposed_layers
for hook in self._propagate_forward_pre_hooks.values():
res = hook(self, (edge_index, size, kwargs))
if res is not None:
edge_index, size, kwargs = res
mutable_size = self._check_input(edge_index, size)
# Run "fused" message and aggregation (if applicable).
fuse = False
if self.fuse and not self.explain:
if is_sparse(edge_index):
fuse = True
elif (not torch.jit.is_scripting()
and isinstance(edge_index, EdgeIndex)):
if (self.SUPPORTS_FUSED_EDGE_INDEX
and edge_index.is_sorted_by_col):
fuse = True
if fuse:
coll_dict = self._collect(self._fused_user_args, edge_index,
mutable_size, kwargs)
msg_aggr_kwargs = self.inspector.collect_param_data(
'message_and_aggregate', coll_dict)
for hook in self._message_and_aggregate_forward_pre_hooks.values():
res = hook(self, (edge_index, msg_aggr_kwargs))
if res is not None:
edge_index, msg_aggr_kwargs = res
out = self.message_and_aggregate(edge_index, **msg_aggr_kwargs)
for hook in self._message_and_aggregate_forward_hooks.values():
res = hook(self, (edge_index, msg_aggr_kwargs), out)
if res is not None:
out = res
update_kwargs = self.inspector.collect_param_data(
'update', coll_dict)
out = self.update(out, **update_kwargs)
else: # Otherwise, run both functions in separation.
if decomposed_layers > 1:
user_args = self._user_args
decomp_args = {a[:-2] for a in user_args if a[-2:] == '_j'}
decomp_kwargs = {
a: kwargs[a].chunk(decomposed_layers, -1)
for a in decomp_args
}
decomp_out = []
for i in range(decomposed_layers):
if decomposed_layers > 1:
for arg in decomp_args:
kwargs[arg] = decomp_kwargs[arg][i]
coll_dict = self._collect(self._user_args, edge_index,
mutable_size, kwargs)
msg_kwargs = self.inspector.collect_param_data(
'message', coll_dict)
for hook in self._message_forward_pre_hooks.values():
res = hook(self, (msg_kwargs, ))
if res is not None:
msg_kwargs = res[0] if isinstance(res, tuple) else res
out = self.message(**msg_kwargs)
for hook in self._message_forward_hooks.values():
res = hook(self, (msg_kwargs, ), out)
if res is not None:
out = res
if self.explain:
explain_msg_kwargs = self.inspector.collect_param_data(
'explain_message', coll_dict)
out = self.explain_message(out, **explain_msg_kwargs)
aggr_kwargs = self.inspector.collect_param_data(
'aggregate', coll_dict)
for hook in self._aggregate_forward_pre_hooks.values():
res = hook(self, (aggr_kwargs, ))
if res is not None:
aggr_kwargs = res[0] if isinstance(res, tuple) else res
out = self.aggregate(out, **aggr_kwargs)
for hook in self._aggregate_forward_hooks.values():
res = hook(self, (aggr_kwargs, ), out)
if res is not None:
out = res
update_kwargs = self.inspector.collect_param_data(
'update', coll_dict)
out = self.update(out, **update_kwargs)
if decomposed_layers > 1:
decomp_out.append(out)
if decomposed_layers > 1:
out = torch.cat(decomp_out, dim=-1)
for hook in self._propagate_forward_hooks.values():
res = hook(self, (edge_index, mutable_size, kwargs), out)
if res is not None:
out = res
return out
[docs] def message(self, x_j: Tensor) -> Tensor:
r"""Constructs messages from node :math:`j` to node :math:`i`
in analogy to :math:`\phi_{\mathbf{\Theta}}` for each edge in
:obj:`edge_index`.
This function can take any argument as input which was initially
passed to :meth:`propagate`.
Furthermore, tensors passed to :meth:`propagate` can be mapped to the
respective nodes :math:`i` and :math:`j` by appending :obj:`_i` or
:obj:`_j` to the variable name, *.e.g.* :obj:`x_i` and :obj:`x_j`.
"""
return x_j
[docs] def aggregate(
self,
inputs: Tensor,
index: Tensor,
ptr: Optional[Tensor] = None,
dim_size: Optional[int] = None,
) -> Tensor:
r"""Aggregates messages from neighbors as
:math:`\bigoplus_{j \in \mathcal{N}(i)}`.
Takes in the output of message computation as first argument and any
argument which was initially passed to :meth:`propagate`.
By default, this function will delegate its call to the underlying
:class:`~torch_geometric.nn.aggr.Aggregation` module to reduce messages
as specified in :meth:`__init__` by the :obj:`aggr` argument.
"""
return self.aggr_module(inputs, index, ptr=ptr, dim_size=dim_size,
dim=self.node_dim)
[docs] @abstractmethod
def message_and_aggregate(self, edge_index: Adj) -> Tensor:
r"""Fuses computations of :func:`message` and :func:`aggregate` into a
single function.
If applicable, this saves both time and memory since messages do not
explicitly need to be materialized.
This function will only gets called in case it is implemented and
propagation takes place based on a :obj:`torch_sparse.SparseTensor`
or a :obj:`torch.sparse.Tensor`.
"""
raise NotImplementedError
[docs] def update(self, inputs: Tensor) -> Tensor:
r"""Updates node embeddings in analogy to
:math:`\gamma_{\mathbf{\Theta}}` for each node
:math:`i \in \mathcal{V}`.
Takes in the output of aggregation as first argument and any argument
which was initially passed to :meth:`propagate`.
"""
return inputs
# Edge-level Updates ######################################################
[docs] def edge_updater(
self,
edge_index: Adj,
size: Size = None,
**kwargs: Any,
) -> Tensor:
r"""The initial call to compute or update features for each edge in the
graph.
Args:
edge_index (torch.Tensor or SparseTensor): A :obj:`torch.Tensor`, a
:class:`torch_sparse.SparseTensor` or a
:class:`torch.sparse.Tensor` that defines the underlying graph
connectivity/message passing flow.
See :meth:`propagate` for more information.
size ((int, int), optional): The size :obj:`(N, M)` of the
assignment matrix in case :obj:`edge_index` is a
:class:`torch.Tensor`.
If set to :obj:`None`, the size will be automatically inferred
and assumed to be quadratic.
This argument is ignored in case :obj:`edge_index` is a
:class:`torch_sparse.SparseTensor` or
a :class:`torch.sparse.Tensor`. (default: :obj:`None`)
**kwargs: Any additional data which is needed to compute or update
features for each edge in the graph.
"""
for hook in self._edge_update_forward_pre_hooks.values():
res = hook(self, (edge_index, size, kwargs))
if res is not None:
edge_index, size, kwargs = res
mutable_size = self._check_input(edge_index, size=None)
coll_dict = self._collect(self._edge_user_args, edge_index,
mutable_size, kwargs)
edge_kwargs = self.inspector.collect_param_data(
'edge_update', coll_dict)
out = self.edge_update(**edge_kwargs)
for hook in self._edge_update_forward_hooks.values():
res = hook(self, (edge_index, size, kwargs), out)
if res is not None:
out = res
return out
[docs] @abstractmethod
def edge_update(self) -> Tensor:
r"""Computes or updates features for each edge in the graph.
This function can take any argument as input which was initially passed
to :meth:`edge_updater`.
Furthermore, tensors passed to :meth:`edge_updater` can be mapped to
the respective nodes :math:`i` and :math:`j` by appending :obj:`_i` or
:obj:`_j` to the variable name, *.e.g.* :obj:`x_i` and :obj:`x_j`.
"""
raise NotImplementedError
# Inference Decomposition #################################################
@property
def decomposed_layers(self) -> int:
return self._decomposed_layers
@decomposed_layers.setter
def decomposed_layers(self, decomposed_layers: int) -> None:
if torch.jit.is_scripting():
raise ValueError("Inference decomposition of message passing "
"modules is only supported on the Python module")
if decomposed_layers == self._decomposed_layers:
return # Abort early if nothing to do.
self._decomposed_layers = decomposed_layers
if decomposed_layers != 1:
if hasattr(self.__class__, '_orig_propagate'):
self.propagate = self.__class__._orig_propagate.__get__(
self, MessagePassing)
elif self.explain is None or self.explain is False:
if hasattr(self.__class__, '_jinja_propagate'):
self.propagate = self.__class__._jinja_propagate.__get__(
self, MessagePassing)
# Explainability ##########################################################
@property
def explain(self) -> Optional[bool]:
return self._explain
@explain.setter
def explain(self, explain: Optional[bool]) -> None:
if torch.jit.is_scripting():
raise ValueError("Explainability of message passing modules "
"is only supported on the Python module")
if explain == self._explain:
return # Abort early if nothing to do.
self._explain = explain
if explain is True:
assert self.decomposed_layers == 1
self.inspector.remove_signature(self.explain_message)
self.inspector.inspect_signature(self.explain_message, exclude=[0])
self._user_args = self.inspector.get_flat_param_names(
funcs=['message', 'explain_message', 'aggregate', 'update'],
exclude=self.special_args,
)
if hasattr(self.__class__, '_orig_propagate'):
self.propagate = self.__class__._orig_propagate.__get__(
self, MessagePassing)
else:
self._user_args = self.inspector.get_flat_param_names(
funcs=['message', 'aggregate', 'update'],
exclude=self.special_args,
)
if self.decomposed_layers == 1:
if hasattr(self.__class__, '_jinja_propagate'):
self.propagate = self.__class__._jinja_propagate.__get__(
self, MessagePassing)
def explain_message(
self,
inputs: Tensor,
dim_size: Optional[int],
) -> Tensor:
# NOTE Replace this method in custom explainers per message-passing
# layer to customize how messages shall be explained, e.g., via:
# conv.explain_message = explain_message.__get__(conv, MessagePassing)
# see stackoverflow.com: 394770/override-a-method-at-instance-level
edge_mask = self._edge_mask
if edge_mask is None:
raise ValueError("Could not find a pre-defined 'edge_mask' "
"to explain. Did you forget to initialize it?")
if self._apply_sigmoid:
edge_mask = edge_mask.sigmoid()
# Some ops add self-loops to `edge_index`. We need to do the same for
# `edge_mask` (but do not train these entries).
if inputs.size(self.node_dim) != edge_mask.size(0):
assert dim_size is not None
edge_mask = edge_mask[self._loop_mask]
loop = edge_mask.new_ones(dim_size)
edge_mask = torch.cat([edge_mask, loop], dim=0)
assert inputs.size(self.node_dim) == edge_mask.size(0)
size = [1] * inputs.dim()
size[self.node_dim] = -1
return inputs * edge_mask.view(size)
# Hooks ###################################################################
[docs] def register_propagate_forward_pre_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward pre-hook on the module.
The hook will be called every time before :meth:`propagate` is invoked.
It should have the following signature:
.. code-block:: python
hook(module, inputs) -> None or modified input
The hook can modify the input.
Input keyword arguments are passed to the hook as a dictionary in
:obj:`inputs[-1]`.
Returns a :class:`torch.utils.hooks.RemovableHandle` that can be used
to remove the added hook by calling :obj:`handle.remove()`.
"""
handle = RemovableHandle(self._propagate_forward_pre_hooks)
self._propagate_forward_pre_hooks[handle.id] = hook
return handle
[docs] def register_propagate_forward_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward hook on the module.
The hook will be called every time after :meth:`propagate` has computed
an output.
It should have the following signature:
.. code-block:: python
hook(module, inputs, output) -> None or modified output
The hook can modify the output.
Input keyword arguments are passed to the hook as a dictionary in
:obj:`inputs[-1]`.
Returns a :class:`torch.utils.hooks.RemovableHandle` that can be used
to remove the added hook by calling :obj:`handle.remove()`.
"""
handle = RemovableHandle(self._propagate_forward_hooks)
self._propagate_forward_hooks[handle.id] = hook
return handle
[docs] def register_message_forward_pre_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward pre-hook on the module.
The hook will be called every time before :meth:`message` is invoked.
See :meth:`register_propagate_forward_pre_hook` for more information.
"""
handle = RemovableHandle(self._message_forward_pre_hooks)
self._message_forward_pre_hooks[handle.id] = hook
return handle
[docs] def register_message_forward_hook(self, hook: Callable) -> RemovableHandle:
r"""Registers a forward hook on the module.
The hook will be called every time after :meth:`message` has computed
an output.
See :meth:`register_propagate_forward_hook` for more information.
"""
handle = RemovableHandle(self._message_forward_hooks)
self._message_forward_hooks[handle.id] = hook
return handle
[docs] def register_aggregate_forward_pre_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward pre-hook on the module.
The hook will be called every time before :meth:`aggregate` is invoked.
See :meth:`register_propagate_forward_pre_hook` for more information.
"""
handle = RemovableHandle(self._aggregate_forward_pre_hooks)
self._aggregate_forward_pre_hooks[handle.id] = hook
return handle
[docs] def register_aggregate_forward_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward hook on the module.
The hook will be called every time after :meth:`aggregate` has computed
an output.
See :meth:`register_propagate_forward_hook` for more information.
"""
handle = RemovableHandle(self._aggregate_forward_hooks)
self._aggregate_forward_hooks[handle.id] = hook
return handle
[docs] def register_message_and_aggregate_forward_pre_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward pre-hook on the module.
The hook will be called every time before :meth:`message_and_aggregate`
is invoked.
See :meth:`register_propagate_forward_pre_hook` for more information.
"""
handle = RemovableHandle(self._message_and_aggregate_forward_pre_hooks)
self._message_and_aggregate_forward_pre_hooks[handle.id] = hook
return handle
[docs] def register_message_and_aggregate_forward_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward hook on the module.
The hook will be called every time after :meth:`message_and_aggregate`
has computed an output.
See :meth:`register_propagate_forward_hook` for more information.
"""
handle = RemovableHandle(self._message_and_aggregate_forward_hooks)
self._message_and_aggregate_forward_hooks[handle.id] = hook
return handle
[docs] def register_edge_update_forward_pre_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward pre-hook on the module.
The hook will be called every time before :meth:`edge_update` is
invoked. See :meth:`register_propagate_forward_pre_hook` for more
information.
"""
handle = RemovableHandle(self._edge_update_forward_pre_hooks)
self._edge_update_forward_pre_hooks[handle.id] = hook
return handle
[docs] def register_edge_update_forward_hook(
self,
hook: Callable,
) -> RemovableHandle:
r"""Registers a forward hook on the module.
The hook will be called every time after :meth:`edge_update` has
computed an output.
See :meth:`register_propagate_forward_hook` for more information.
"""
handle = RemovableHandle(self._edge_update_forward_hooks)
self._edge_update_forward_hooks[handle.id] = hook
return handle
# TorchScript Support #####################################################
def _set_jittable_templates(self, raise_on_error: bool = False) -> None:
root_dir = osp.dirname(osp.realpath(__file__))
jinja_prefix = f'{self.__module__}_{self.__class__.__name__}'
# Optimize `propagate()` via `*.jinja` templates:
if not self.propagate.__module__.startswith(jinja_prefix):
try:
if ('propagate' in self.__class__.__dict__
and self.__class__.__dict__['propagate']
!= MessagePassing.propagate):
raise ValueError("Cannot compile custom 'propagate' "
"method")
module = module_from_template(
module_name=f'{jinja_prefix}_propagate',
template_path=osp.join(root_dir, 'propagate.jinja'),
tmp_dirname='message_passing',
# Keyword arguments:
modules=self.inspector._modules,
collect_name='collect',
signature=self._get_propagate_signature(),
collect_param_dict=self.inspector.get_flat_param_dict(
['message', 'aggregate', 'update']),
message_args=self.inspector.get_param_names('message'),
aggregate_args=self.inspector.get_param_names('aggregate'),
message_and_aggregate_args=self.inspector.get_param_names(
'message_and_aggregate'),
update_args=self.inspector.get_param_names('update'),
fuse=self.fuse,
)
self.__class__._orig_propagate = self.__class__.propagate
self.__class__._jinja_propagate = module.propagate
self.__class__.propagate = module.propagate
self.__class__.collect = module.collect
except Exception as e: # pragma: no cover
if raise_on_error:
raise e
self.__class__._orig_propagate = self.__class__.propagate
self.__class__._jinja_propagate = self.__class__.propagate
# Optimize `edge_updater()` via `*.jinja` templates (if implemented):
if (self.inspector.implements('edge_update')
and not self.edge_updater.__module__.startswith(jinja_prefix)):
try:
if ('edge_updater' in self.__class__.__dict__
and self.__class__.__dict__['edge_updater']
!= MessagePassing.edge_updater):
raise ValueError("Cannot compile custom 'edge_updater' "
"method")
module = module_from_template(
module_name=f'{jinja_prefix}_edge_updater',
template_path=osp.join(root_dir, 'edge_updater.jinja'),
tmp_dirname='message_passing',
# Keyword arguments:
modules=self.inspector._modules,
collect_name='edge_collect',
signature=self._get_edge_updater_signature(),
collect_param_dict=self.inspector.get_param_dict(
'edge_update'),
)
self.__class__._orig_edge_updater = self.__class__.edge_updater
self.__class__._jinja_edge_updater = module.edge_updater
self.__class__.edge_updater = module.edge_updater
self.__class__.edge_collect = module.edge_collect
except Exception as e: # pragma: no cover
if raise_on_error:
raise e
self.__class__._orig_edge_updater = self.__class__.edge_updater
self.__class__._jinja_edge_updater = (
self.__class__.edge_updater)
def _get_propagate_signature(self) -> Signature:
param_dict = self.inspector.get_params_from_method_call(
'propagate', exclude=[0, 'edge_index', 'size'])
update_signature = self.inspector.get_signature('update')
return Signature(
param_dict=param_dict,
return_type=update_signature.return_type,
return_type_repr=update_signature.return_type_repr,
)
def _get_edge_updater_signature(self) -> Signature:
param_dict = self.inspector.get_params_from_method_call(
'edge_updater', exclude=[0, 'edge_index', 'size'])
edge_update_signature = self.inspector.get_signature('edge_update')
return Signature(
param_dict=param_dict,
return_type=edge_update_signature.return_type,
return_type_repr=edge_update_signature.return_type_repr,
)
[docs] def jittable(self, typing: Optional[str] = None) -> 'MessagePassing':
r"""Analyzes the :class:`MessagePassing` instance and produces a new
jittable module that can be used in combination with
:meth:`torch.jit.script`.
.. note::
:meth:`jittable` is deprecated and a no-op from :pyg:`PyG` 2.5
onwards.
"""
warnings.warn(f"'{self.__class__.__name__}.jittable' is deprecated "
f"and a no-op. Please remove its usage.")
return self