MessagePassing

1.相关包导入

import os
import re
import inspect
import os.path as osp
from uuid import uuid1
from itertools import chain
from inspect import Parameter
from typing import List, Optional, Set
from torch_geometric.typing import Adj, Size

import torch
from torch import Tensor
from jinja2 import Template
from torch_sparse import SparseTensor
from torch_scatter import gather_csr, scatter, segment_csr

from .utils.helpers import expand_left
from .utils.jit import class_from_module_repr
from .utils.typing import (sanitize, split_types_repr, parse_types,
                           resolve_types)
from .utils.inspector import Inspector, func_header_repr, func_body_repr

2.

2.1 数学表达

${\mathbf{x}}_i^{(k)}={\gamma }^{(k)}\left({\mathbf{x}}_i^{(k-1)},{\square }_{j\in \mathcal{N}(i)}{\varphi }^{(k)}\left({\mathbf{x}}_i^{(k-1)},{\mathbf{x}}_j^{(k-1)},{\mathbf{e}}_{j,i}\right)\right),$

$\square$ d表示可微的置换不变函数, e.g., sum, mean or max
$\gamma$ and $\varphi$ 表示可微函数，例如 MLP（多层感知器）.

2.2三个用户自定义函数

上述三个函数是需要需用户自定义的，即；
1.$\varphi$-message()
2.$\gamma$ -update(),
3.$\square -$ 聚合方式, aggr=“add”,aggr="mean"或aggr=“max”

3 源码

3.1 MessagePassing(aggr=“add”, flow=“source_to_target”, node_dim=-2)

  aggr (string, optional): 聚合方案 ( "add","mean"或"max") 
      (default: :`"add"`)
  flow (string, optional): 消息传递的流向 ("source_to_target"或"target_to_source") 也就是A→B还是B→A