04_DGL_01_框架介绍.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "f0cf0120",
   "metadata": {},
   "source": [
    "### 一.运行机制\n",
    "为了更加高效的运行GNN，我们可以利用一些高效的图计算框架帮助我们训练模型，DGL（Deep Graph Library）便是一种比较方便的框架，它将所有的图计算过程拆分为了三个基本部分，分别介绍如下：   \n",
    "\n",
    "**1.1 消息传递函数**  \n",
    "该函数是将一条边以及与其关联的两个节点的信息进行聚合，然后将聚合后**消息**重新赋值到边上，可以用如下的表达式抽象表达：   \n",
    "\n",
    "$$\n",
    "m_{uv}^{(t+1)}=\\phi(x_u^{(t)},x_v^{(t)},w_{uv}^{(t)})\n",
    "$$  \n",
    "\n",
    "其中，$w_{uv}^{(t)}$表示边$(u,v)$上的特征，$x_u^{(t)}$表示节点$u$上的特征，$x_v^{(t)}$表示节点$v$上的特征  \n",
    "\n",
    "**1.2 聚合函数**   \n",
    "聚合函数是将与训练节点关联的边上的消息，进行聚合：    \n",
    "\n",
    "$$\n",
    "\\rho_u^{t+1}=\\rho(\\{m_{uk}^{(t+1)}\\mid k\\in N(u)\\})\n",
    "$$  \n",
    "\n",
    "这里，$N(u)$表示与节点$u$相连的邻居节点   \n",
    "\n",
    "**1.3 更新函数**  \n",
    "更新函数的作用是将聚合特征与节点的旧特征进行聚合，然后生成节点的新特征：   \n",
    "\n",
    "$$\n",
    "x_u^{(t+1)}=\\varphi(x_u^{(t)},\\rho_u^{t+1})\n",
    "$$   \n",
    "\n",
    "我们之前介绍的GCN，GAT，GraphSAGE便可分解为这三个基本操作函数，接下来我利用DGL库演示一下这3个基本操作"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f4785f64",
   "metadata": {},
   "source": [
    "### 二.实践  \n",
    "\n",
    "**2.1 构图**：首先构建一张图，并为节点和边赋值"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "c6897da0",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using backend: pytorch\n"
     ]
    }
   ],
   "source": [
    "import dgl\n",
    "import torch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "bd0170fb",
   "metadata": {},
   "outputs": [],
   "source": [
    "u=torch.tensor([0,0,0,0,1,1,2,3,2,4,5,5,6,6,6,7,8])\n",
    "v=torch.tensor([1,2,3,4,2,3,4,4,5,5,6,7,7,8,9,8,9])\n",
    "g=dgl.graph((u,v))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "17ced9cd",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#转换为networkx可视化\n",
    "%matplotlib inline\n",
    "from matplotlib import pyplot as plt\n",
    "import networkx as nx\n",
    "plt.axis(\"off\")\n",
    "nx.draw_networkx(g.to_networkx())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "278828b6",
   "metadata": {},
   "outputs": [],
   "source": [
    "#分别为节点和边赋值一个三维的向量\n",
    "g.ndata[\"nx\"]=torch.randn(g.num_nodes(),3)\n",
    "g.edata[\"ex\"]=torch.randn(g.num_edges(),3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "596a931a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(tensor([[ 0.6239,  0.3987, -0.6238],\n",
       "         [-0.7141, -1.0818, -1.0757],\n",
       "         [-0.0540,  1.1078,  0.2700],\n",
       "         [ 0.5681, -1.0023, -1.4963],\n",
       "         [-1.3526, -0.5072,  0.3111],\n",
       "         [ 0.4039,  0.1909,  2.4836],\n",
       "         [ 0.2190,  0.9467, -0.6446],\n",
       "         [-0.5860, -1.7527, -1.0872],\n",
       "         [ 0.8450,  0.9458, -0.5495],\n",
       "         [ 1.3140,  0.4514,  0.5442]]),\n",
       " tensor([[-0.5180, -0.1497,  0.4278],\n",
       "         [ 0.7970, -0.1222, -0.5810],\n",
       "         [-0.9885,  0.7971, -1.0056],\n",
       "         [-0.5377,  0.3229, -0.1410],\n",
       "         [ 1.2062,  1.3272, -0.4768],\n",
       "         [-0.6856,  0.9685, -1.3733],\n",
       "         [ 1.0140, -0.1059,  1.4300],\n",
       "         [-0.4088,  0.5824, -0.4151],\n",
       "         [ 1.4490, -1.0892,  0.2116],\n",
       "         [ 0.2986, -0.0652, -0.4363],\n",
       "         [ 0.7020,  0.6645, -1.1004],\n",
       "         [ 0.7326,  0.5491,  0.8222],\n",
       "         [-0.5815, -1.3774,  0.7640],\n",
       "         [ 1.1323, -1.4232,  0.2073],\n",
       "         [-0.3632, -1.3108, -1.2507],\n",
       "         [-0.7597, -1.1140,  0.2626],\n",
       "         [-0.8476, -1.2887, -0.7922]]))"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#查看\n",
    "g.ndata[\"nx\"],g.edata[\"ex\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "69e84cce",
   "metadata": {},
   "source": [
    "**2.2 利用apply_edges进行消息传递**  \n",
    "\n",
    "apply_edges可以完成第一个函数的操作，即将两节点特征和边特征进行某种操作，然后将结果保存回边上，我们先用dgl自带的api，将$u,v$节点的特征相加然后赋值到边上"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "11fed676",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-0.0902, -0.6831, -1.6995],\n",
       "        [ 0.5699,  1.5065, -0.3538],\n",
       "        [ 1.1919, -0.6036, -2.1202],\n",
       "        [-0.7288, -0.1085, -0.3128],\n",
       "        [-0.7681,  0.0260, -0.8057],\n",
       "        [-0.1461, -2.0841, -2.5721],\n",
       "        [-1.4066,  0.6007,  0.5811],\n",
       "        [-0.7846, -1.5095, -1.1853],\n",
       "        [ 0.3500,  1.2987,  2.7536],\n",
       "        [-0.9487, -0.3163,  2.7946],\n",
       "        [ 0.6229,  1.1375,  1.8389],\n",
       "        [-0.1821, -1.5618,  1.3964],\n",
       "        [-0.3670, -0.8060, -1.7318],\n",
       "        [ 1.0640,  1.8925, -1.1941],\n",
       "        [ 1.5330,  1.3981, -0.1005],\n",
       "        [ 0.2589, -0.8069, -1.6367],\n",
       "        [ 2.1590,  1.3972, -0.0053]])"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g.apply_edges(dgl.function.u_add_v(\"nx\",\"nx\",\"add_x\"))#第一个nx表示u上的nx特征，第二个表示v上的nx特征，add_x表示赋值到边上的特征\n",
    "g.edata[\"add_x\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "be0091a8",
   "metadata": {},
   "source": [
    "这个过程也可以写自定义函数进行操作"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "b297ca46",
   "metadata": {},
   "outputs": [],
   "source": [
    "def message_func(edges):\n",
    "     return {'new_add_x': edges.src['nx'] + edges.dst['nx']}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "828c5b57",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[-0.0902, -0.6831, -1.6995],\n",
       "        [ 0.5699,  1.5065, -0.3538],\n",
       "        [ 1.1919, -0.6036, -2.1202],\n",
       "        [-0.7288, -0.1085, -0.3128],\n",
       "        [-0.7681,  0.0260, -0.8057],\n",
       "        [-0.1461, -2.0841, -2.5721],\n",
       "        [-1.4066,  0.6007,  0.5811],\n",
       "        [-0.7846, -1.5095, -1.1853],\n",
       "        [ 0.3500,  1.2987,  2.7536],\n",
       "        [-0.9487, -0.3163,  2.7946],\n",
       "        [ 0.6229,  1.1375,  1.8389],\n",
       "        [-0.1821, -1.5618,  1.3964],\n",
       "        [-0.3670, -0.8060, -1.7318],\n",
       "        [ 1.0640,  1.8925, -1.1941],\n",
       "        [ 1.5330,  1.3981, -0.1005],\n",
       "        [ 0.2589, -0.8069, -1.6367],\n",
       "        [ 2.1590,  1.3972, -0.0053]])"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g.apply_edges(message_func)\n",
    "g.edata[\"new_add_x\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e4419022",
   "metadata": {},
   "source": [
    "**2.3 利用update_all组合消息传递函数与聚合函数**  \n",
    "\n",
    "update_all将消息传递和聚合函数一并进行操作，下面演示先将邻居节点特征加到边，然后再将边特征加到节点"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "af2e0e50",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 0.0000,  0.0000,  0.0000],\n",
       "        [-0.0902, -0.6831, -1.6995],\n",
       "        [-0.1982,  1.5326, -1.1595],\n",
       "        [ 1.0459, -2.6876, -4.6922],\n",
       "        [-2.9200, -1.0173, -0.9170],\n",
       "        [-0.5987,  0.9824,  5.5482],\n",
       "        [ 0.6229,  1.1375,  1.8389],\n",
       "        [-0.5491, -2.3679, -0.3355],\n",
       "        [ 1.3229,  1.0856, -2.8309],\n",
       "        [ 3.6919,  2.7953, -0.1058]])"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g.update_all(dgl.function.u_add_v(\"nx\",\"nx\",\"add_x\"),\n",
    "             dgl.function.sum(\"add_x\",\"add_edge_x\"))\n",
    "g.ndata[\"add_edge_x\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3d09ae1f",
   "metadata": {},
   "source": [
    "当然，也可以自定义"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "4ffafb68",
   "metadata": {},
   "outputs": [],
   "source": [
    "def reduce_func(nodes):\n",
    "     return {'new_add_edge_x': torch.sum(nodes.mailbox['add_x'], dim=1)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "be5a67b7",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 0.0000,  0.0000,  0.0000],\n",
       "        [-0.0902, -0.6831, -1.6995],\n",
       "        [-0.1982,  1.5326, -1.1595],\n",
       "        [ 1.0459, -2.6876, -4.6922],\n",
       "        [-2.9200, -1.0173, -0.9170],\n",
       "        [-0.5987,  0.9824,  5.5482],\n",
       "        [ 0.6229,  1.1375,  1.8389],\n",
       "        [-0.5491, -2.3679, -0.3355],\n",
       "        [ 1.3229,  1.0856, -2.8309],\n",
       "        [ 3.6919,  2.7953, -0.1058]])"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g.update_all(dgl.function.u_add_v(\"nx\",\"nx\",\"add_x\"),\n",
    "             reduce_func)\n",
    "g.ndata[\"new_add_edge_x\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "36cdfce7",
   "metadata": {},
   "source": [
    "**2.3 更新函数：最后组合聚合特征和原特征**   \n",
    "这部分操作就很容易了，因为聚合特征和原特征都在节点上了，所以我们直接操作即可，比如相加"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "b01f0bce",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 0.6239,  0.3987, -0.6238],\n",
       "        [-0.7141, -1.0818, -1.0757],\n",
       "        [-0.0540,  1.1078,  0.2700],\n",
       "        [ 0.5681, -1.0023, -1.4963],\n",
       "        [-1.3526, -0.5072,  0.3111],\n",
       "        [ 0.4039,  0.1909,  2.4836],\n",
       "        [ 0.2190,  0.9467, -0.6446],\n",
       "        [-0.5860, -1.7527, -1.0872],\n",
       "        [ 0.8450,  0.9458, -0.5495],\n",
       "        [ 1.3140,  0.4514,  0.5442]])"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g.ndata[\"nx\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "31add3a2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[ 0.6239,  0.3987, -0.6238],\n",
       "        [-0.8044, -1.7649, -2.7753],\n",
       "        [-0.2522,  2.6404, -0.8895],\n",
       "        [ 1.6139, -3.6899, -6.1886],\n",
       "        [-4.2726, -1.5244, -0.6059],\n",
       "        [-0.1948,  1.1733,  8.0318],\n",
       "        [ 0.8419,  2.0842,  1.1943],\n",
       "        [-1.1352, -4.1206, -1.4227],\n",
       "        [ 2.1679,  2.0314, -3.3804],\n",
       "        [ 5.0059,  3.2467,  0.4384]])"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "g.ndata[\"nx\"]=g.ndata[\"nx\"]+g.ndata[\"add_edge_x\"]\n",
    "g.ndata[\"nx\"]#注意从第二行开始看"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4e5b6efd",
   "metadata": {},
   "source": [
    "下一节，让我们用PageRank来练练手~"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bc4dcaba",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}