CS224W_Colab_3.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "XuXWJLEm2UWS"
   },
   "source": [
    "# **CS224W - Colab 3**"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "8gzsP50bF6Gb"
   },
   "source": [
    "In Colab 2 we constructed GNN models by using PyTorch Geometric built in GCN layer, the `GCNConv`. In this Colab we will implement the **GraphSAGE** ([Hamilton et al. (2017)](https://arxiv.org/abs/1706.02216)) and **GAT** ([Veličković et al. (2018)](https://arxiv.org/abs/1710.10903)) layers directly. Then we will run our models on the CORA dataset, which is a standard citation network benchmark dataset.\n",
    "\n",
    "We will then use [DeepSNAP](https://snap.stanford.edu/deepsnap/), a Python library assisting efficient deep learning on graphs, to split the graphs in different settings and apply dataset transformations.\n",
    "\n",
    "At last, using DeepSNAP transductive link prediction split functionality, we will construct a simple GNN model on the edge property predition (link prediction) task.\n",
    "\n",
    "**Note**: Make sure to **sequentially run all the cells in each section**, so that the intermediate variables / packages will carry over to the next cell\n",
    "\n",
    "Have fun on Colab 3 :)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "MSaetj53YnT6"
   },
   "source": [
    "# Device\n",
    "You might need to use GPU for this Colab.\n",
    "\n",
    "Please click `Runtime` and then `Change runtime type`. Then set the `hardware accelerator` to **GPU**."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "67gOQITlCNQi"
   },
   "source": [
    "## Installation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "id": "PRfgbfTjCRD_"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'1.7.0'"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import torch_geometric\n",
    "torch_geometric.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "ZoXlf4MtYrbz"
   },
   "source": [
    "# 1 GNN Layers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "TQy2RBfgYut4"
   },
   "source": [
    "## Implementing Layer Modules\n",
    "\n",
    "In colab 2, we implemented a network using GCN in node and graph classification tasks. However, the GCN module we used in colab 2 is from the official library. For this problem, we will provide you with a general Graph Neural Network Stack, where you'll be able to plugin your own modules of GraphSAGE and GATs. We will use our implementations to complete node classification on CORA, which is a standard citation network benchmark dataset. In this dataset, nodes correspond to documents and edges correspond to undirected citations. Each node has a class label. The node features are elements of a bag-or-words representation of a document. For the Cora dataset, there are 2708 nodes, 5429 edges, 7 prediction classes for nodes, and 1433 features per node. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "H4ne6Gw-CT5G"
   },
   "source": [
    "## GNN Stack Module\n",
    "\n",
    "Below is the implementation for a general GNN Module that could plugin any layers, including **GraphSage**, **GAT**, etc. This module is provided for you, and you own **GraphSage** and **GAT** layers will function as components in the GNNStack Module."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "id": "Ys8vZAFPCWWe"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch_scatter\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "\n",
    "import torch_geometric.nn as pyg_nn\n",
    "import torch_geometric.utils as pyg_utils\n",
    "\n",
    "from torch import Tensor\n",
    "from typing import Union, Tuple, Optional\n",
    "from torch_geometric.typing import (OptPairTensor, Adj, Size, NoneType,\n",
    "                                    OptTensor)\n",
    "\n",
    "from torch.nn import Parameter, Linear\n",
    "from torch_sparse import SparseTensor, set_diag\n",
    "from torch_geometric.nn.conv import MessagePassing\n",
    "from torch_geometric.utils import remove_self_loops, add_self_loops, softmax\n",
    "\n",
    "class GNNStack(torch.nn.Module):\n",
    "    def __init__(self, input_dim, hidden_dim, output_dim, args, emb=False):\n",
    "        super(GNNStack, self).__init__()\n",
    "        conv_model = self.build_conv_model(args.model_type)\n",
    "        self.convs = nn.ModuleList()\n",
    "        self.convs.append(conv_model(input_dim, hidden_dim, heads=args.heads))\n",
    "        assert (args.num_layers >= 1), 'Number of layers is not >=1'\n",
    "        for l in range(args.num_layers-1):\n",
    "            self.convs.append(conv_model(args.heads * hidden_dim, hidden_dim, \n",
    "                                         heads=args.heads))\n",
    "\n",
    "        # post-message-passing\n",
    "        self.post_mp = nn.Sequential(\n",
    "            nn.Linear(args.heads * hidden_dim, hidden_dim), nn.Dropout(args.dropout), \n",
    "            nn.Linear(hidden_dim, output_dim))\n",
    "\n",
    "        self.dropout = args.dropout\n",
    "        self.num_layers = args.num_layers\n",
    "\n",
    "        self.emb = emb\n",
    "\n",
    "    def build_conv_model(self, model_type):\n",
    "        if model_type == 'GraphSage':\n",
    "            return GraphSage\n",
    "        elif model_type == 'GAT':\n",
    "            # When applying GAT with num heads > 1, one needs to modify the \n",
    "            # input and output dimension of the conv layers (self.convs),\n",
    "            # to ensure that the input dim of the next layer is num heads\n",
    "            # multiplied by the output dim of the previous layer.\n",
    "            # HINT: In case you want to play with multiheads, you need to change the for-loop when builds up self.convs to be\n",
    "            # self.convs.append(conv_model(hidden_dim * num_heads, hidden_dim)), \n",
    "            # and also the first nn.Linear(hidden_dim * num_heads, hidden_dim) in post-message-passing.\n",
    "            return GAT\n",
    "\n",
    "    def forward(self, data):\n",
    "        x, edge_index, batch = data.x, data.edge_index, data.batch\n",
    "          \n",
    "        for i in range(self.num_layers):\n",
    "            x = self.convs[i](x, edge_index)\n",
    "            x = F.relu(x)\n",
    "            x = F.dropout(x, p=self.dropout)\n",
    "\n",
    "        x = self.post_mp(x)\n",
    "\n",
    "        if self.emb == True:\n",
    "            return x\n",
    "\n",
    "        return F.log_softmax(x, dim=1)\n",
    "\n",
    "    def loss(self, pred, label):\n",
    "        return F.nll_loss(pred, label)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "syDtxjxoCZgq"
   },
   "source": [
    "## GraphSage Implementation\n",
    "\n",
    "Now let's start working on our own implementation of layers! This part is to get you familiar with how to implement Pytorch layer based on Message Passing. You will be implementing the **forward**, **message** and **aggregate** functions.\n",
    "\n",
    "Generally, the **forward** function is where the actual message passing is conducted. All logic in each iteration happens in **forward**, where we'll call **propagate** function to propagate information from neighbor nodes to central nodes.  So the general paradigm will be pre-processing -> propagate -> post-processing.\n",
    "\n",
    "Recall the process of message passing we introduced in homework 1. **propagate** further calls **message** which transforms information of neighbor nodes into messages, **aggregate** which aggregates all messages from neighbor nodes into one, and **update** which further generates the embedding for nodes in the next iteration.\n",
    "\n",
    "Our implementation is slightly variant from this, where we'll not explicitly implement **update**, but put the logic for updating nodes in **forward** function. To be more specific, after information is propagated, we can further conduct some operations on the output of **propagate**. The output of **forward** is exactly the embeddings after the current iteration.\n",
    "\n",
    "In addition, tensors passed to **propagate()** can be mapped to the respective nodes $i$ and $j$ by appending _i or _j to the variable name, .e.g. x_i and x_j. Note that we generally refer to $i$ as the central nodes that aggregates information, and refer to $j$ as the neighboring nodes, since this is the most common notation.\n",
    "\n",
    "Please find more details in the comments. One thing to note is that we're adding **skip connections** to our GraphSage. Formally, the update rule for our model is described as below:\n",
    "\n",
    "\\begin{equation}\n",
    "h_v^{(l)} = W_l\\cdot h_v^{(l-1)} + W_r \\cdot AGG(\\{h_u^{(l-1)}, \\forall u \\in N(v) \\})\n",
    "\\end{equation}\n",
    "\n",
    "For simplicity, we use mean aggregations where:\n",
    "\n",
    "\\begin{equation}\n",
    "AGG(\\{h_u^{(l-1)}, \\forall u \\in N(v) \\}) = \\frac{1}{|N(v)|} \\sum_{u\\in N(v)} h_u^{(l-1)}\n",
    "\\end{equation}\n",
    "\n",
    "Additionally, $\\ell$-2 normalization is applied after each iteration.\n",
    "\n",
    "In order to complete the work correctly, we have to understand how the different functions interact with each other. In **propagate** we can pass in any parameters we want. For example, we pass in $x$ as an parameter:\n",
    "\n",
    "... = propagate(..., $x$=($x_{central}$, $x_{neighbor}$), ...)\n",
    "\n",
    "Here $x_{central}$ and $x_{neighbor}$ represent the features from **central** nodes and from **neighbor** nodes. If we're using the same representations from central and neighbor, then $x_{central}$ and $x_{neighbor}$ could be identical.\n",
    "\n",
    "Suppose $x_{central}$ and $x_{neighbor}$ are both of shape N * d, where N is number of nodes, and d is dimension of features.\n",
    "\n",
    "Then in message function, we can take parameters called $x\\_i$ and $x\\_j$. Usually $x\\_i$ represents \"central nodes\", and $x\\_j$ represents \"neighbor nodes\". Pay attention to the shape here: $x\\_i$ and $x\\_j$ are both of shape E * d (**not N!**). $x\\_i$ is obtained by concatenating the embeddings of central nodes of all edges through lookups from $x_{central}$ we passed in propagate. Similarly, $x\\_j$ is obtained by concatenating the embeddings of neighbor nodes of all edges through lookups from $x_{neighbor}$ we passed in propagate.\n",
    "\n",
    "Let's look at an example. Suppose we have 4 nodes, so $x_{central}$ and $x_{neighbor}$ are of shape 4 * d. We have two edges (1, 2) and (3, 0). Thus, $x\\_i$ is obtained by $[x_{central}[1]^T; x_{central}[3]^T]^T$, and $x\\_j$ is obtained by $[x_{neighbor}[2]^T; x_{neighbor}[0]^T]^T$\n",
    "\n",
    "<font color='red'>For the following questions, DON'T refer to any existing implementations online.</font>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "RwG4HqCFCaOD"
   },
   "outputs": [],
   "source": [
    "class GraphSage(MessagePassing):\n",
    "    \n",
    "    def __init__(self, in_channels, out_channels, normalize = True,\n",
    "                 bias = False, heads=1, **kwargs):  \n",
    "        super(GraphSage, self).__init__(**kwargs)\n",
    "\n",
    "        self.in_channels = in_channels\n",
    "        self.out_channels = out_channels\n",
    "        self.normalize = normalize\n",
    "\n",
    "        self.lin_l = None\n",
    "        self.lin_r = None\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Define the layers needed for the message and update functions below.\n",
    "        # self.lin_l is the linear transformation that you apply to embedding \n",
    "        #            for central node.\n",
    "        # self.lin_r is the linear transformation that you apply to aggregated \n",
    "        #            message from neighbors.\n",
    "        # Our implementation is ~2 lines, but don't worry if you deviate from this.\n",
    "        self.lin_l = nn.Linear(self.in_channels, self.out_channels)\n",
    "        self.lin_r = nn.Linear(self.in_channels, self.out_channels)\n",
    "\n",
    "        ############################################################################\n",
    "\n",
    "        self.reset_parameters()\n",
    "\n",
    "    def reset_parameters(self):\n",
    "        self.lin_l.reset_parameters()\n",
    "        self.lin_r.reset_parameters()\n",
    "\n",
    "    def forward(self, x, edge_index, size = None):\n",
    "        \"\"\"\"\"\"\n",
    "\n",
    "        out = None\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Implement message passing, as well as any post-processing (our update rule).\n",
    "        # 1. First call propagate function to conduct the message passing.\n",
    "        #    1.1 See there for more information: \n",
    "        #        https://pytorch-geometric.readthedocs.io/en/latest/notes/create_gnn.html\n",
    "        #    1.2 We use the same representations for central (x_central) and \n",
    "        #        neighbor (x_neighbor) nodes, which means you'll pass x=(x, x) \n",
    "        #        to propagate.\n",
    "        # 2. Update our node embedding with skip connection.\n",
    "        # 3. If normalize is set, do L-2 normalization (defined in \n",
    "        #    torch.nn.functional)\n",
    "        # Our implementation is ~5 lines, but don't worry if you deviate from this.\n",
    "        x_prop = self.propagate(edge_index, x=(x,x), size=size)\n",
    "        x = self.lin_l(x) + x_prop\n",
    "        \n",
    "        if self.normalize:\n",
    "            x = F.normalize(x)\n",
    "        \n",
    "        out = x\n",
    "        \n",
    "        ############################################################################\n",
    "\n",
    "        return out\n",
    "\n",
    "    def message(self, x_j):\n",
    "\n",
    "        out = None\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Implement your message function here.\n",
    "        # Our implementation is ~1 lines, but don't worry if you deviate from this.\n",
    "        out = self.lin_r(x_j)\n",
    "\n",
    "        ############################################################################\n",
    "\n",
    "        return out\n",
    "\n",
    "    def aggregate(self, inputs, index, dim_size = None):\n",
    "\n",
    "        out = None\n",
    "\n",
    "        # The axis along which to index number of nodes.\n",
    "        node_dim = self.node_dim\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Implement your aggregate function here.\n",
    "        # See here as how to use torch_scatter.scatter: \n",
    "        # https://pytorch-scatter.readthedocs.io/en/latest/functions/scatter.html#torch_scatter.scatter\n",
    "        # Our implementation is ~1 lines, but don't worry if you deviate from this.\n",
    "        out = torch_scatter.scatter(inputs, index, dim=node_dim, reduce=\"mean\")\n",
    "\n",
    "        ############################################################################\n",
    "\n",
    "        return out\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "qjcfF3RACdLD"
   },
   "source": [
    "## GAT Implementation\n",
    "\n",
    "Attention mechanisms have become the state-of-the-art in many sequence-based tasks such as machine translation and learning sentence representations. One of the major benefits of attention-based mechanisms is their ability to focus on the most relevant parts of the input to make decisions. In this problem, we will see how attention mechanisms can be used to perform node classification of graph-structured data through the usage of Graph Attention Networks (GATs).\n",
    "\n",
    "The building block of the Graph Attention Network is the graph attention layer, which is a variant of the aggregation function . Let $N$ be the number of nodes and $F$ be the dimension of the feature vector for each node. The input to each graph attentional layer is a set of node features: $\\mathbf{h} = \\{\\overrightarrow{h_1}, \\overrightarrow{h_2}, \\dots, \\overrightarrow{h_N}$\\}, $\\overrightarrow{h_i} \\in R^F$. The output of each graph attentional layer is a new set of node features, which may have a new dimension $F'$: $\\mathbf{h'} = \\{\\overrightarrow{h_1'}, \\overrightarrow{h_2'}, \\dots, \\overrightarrow{h_N'}\\}$, with $\\overrightarrow{h_i'} \\in \\mathbb{R}^{F'}$.\n",
    "\n",
    "We will now describe this transformation of the input features into higher-level features performed by each graph attention layer. First, a shared linear transformation parametrized by the weight matrix $\\mathbf{W} \\in \\mathbb{R}^{F' \\times F}$ is applied to every node. Next, we perform self-attention on the nodes. We use a shared attentional mechanism:\n",
    "\\begin{equation} \n",
    "a : \\mathbb{R}^{F'} \\times \\mathbb{R}^{F'} \\rightarrow \\mathbb{R}.\n",
    "\\end{equation}\n",
    "\n",
    "This mechanism computes the attention coefficients that capture the importance of node $j$'s features to node $i$:\n",
    "\\begin{equation}\n",
    "e_{ij} = a(\\mathbf{W_l}\\overrightarrow{h_i}, \\mathbf{W_r} \\overrightarrow{h_j})\n",
    "\\end{equation}\n",
    "The most general formulation of self-attention allows every node to attend to all other nodes which drops all structural information. To utilize graph structure in the attention mechanisms, we can use masked attention. In masked attention, we only compute $e_{ij}$ for nodes $j \\in \\mathcal{N}_i$ where $\\mathcal{N}_i$ is some neighborhood of node $i$ in the graph.\n",
    "\n",
    "To easily compare coefficients across different nodes, we normalize the coefficients across $j$ using a softmax function:\n",
    "\\begin{equation}\n",
    "\\alpha_{ij} = \\text{softmax}_j(e_{ij}) = \\frac{\\exp(e_{ij})}{\\sum_{k \\in \\mathcal{N}_i} \\exp(e_{ik})}\n",
    "\\end{equation}\n",
    "\n",
    "For this problem, our attention mechanism $a$ will be a single-layer feedforward neural network parametrized by a weight vector $\\overrightarrow{a} \\in \\mathbb{R}^{F'}$, followed by a LeakyReLU nonlinearity (with negative input slope 0.2). Let $\\cdot^T$ represent transposition and $||$ represent concatenation. The coefficients computed by our attention mechanism may be expressed as:\n",
    "\n",
    "\\begin{equation}\n",
    "\\alpha_{ij} = \\frac{\\exp\\Big(\\text{LeakyReLU}\\Big(\\overrightarrow{a_l}^T \\mathbf{W_l} \\overrightarrow{h_i} + \\overrightarrow{a_r}^T\\mathbf{W_r}\\overrightarrow{h_j}\\Big)\\Big)}{\\sum_{k\\in \\mathcal{N}_i} \\exp\\Big(\\text{LeakyReLU}\\Big(\\overrightarrow{a_l}^T \\mathbf{W_l} \\overrightarrow{h_i} + \\overrightarrow{a_r}^T\\mathbf{W_r}\\overrightarrow{h_k}\\Big)\\Big)}\n",
    "\\end{equation}\n",
    "\n",
    "For the following questions, we denote $\\alpha_l = [...,\\overrightarrow{a_l}^T \\mathbf{W_l} \\overrightarrow{h_i},...]$ and $\\alpha_r = [..., \\overrightarrow{a_r}^T \\mathbf{W_r} \\overrightarrow{h_j}, ...]$.\n",
    "\n",
    "\n",
    "At every layer of GAT, after the attention coefficients are computed for that layer, the aggregation function can be computed by a weighted sum of neighborhood messages, where weights are specified by $\\alpha_{ij}$.\n",
    "\n",
    "Now, we use the normalized attention coefficients to compute a linear combination of the features corresponding to them. These aggregated features will serve as the final output features for every node.\n",
    "\n",
    "\\begin{equation}\n",
    "h_i' = \\sum_{j \\in \\mathcal{N}_i} \\alpha_{ij} \\mathbf{W_r} \\overrightarrow{h_j}.\n",
    "\\end{equation}\n",
    "\n",
    "To stabilize the learning process of self-attention, we use multi-head attention. To do this we use $K$ independent attention mechanisms, or ``heads'' compute output features as in the above equations. Then, we concatenate these output feature representations:\n",
    "\n",
    "\\begin{equation}\n",
    "    \\overrightarrow{h_i}' = ||_{k=1}^K \\Big(\\sum_{j \\in \\mathcal{N}_i} \\alpha_{ij}^{(k)} \\mathbf{W_r}^{(k)} \\overrightarrow{h_j}\\Big)\n",
    "\\end{equation}\n",
    "\n",
    "where $||$ is concentation, $\\alpha_{ij}^{(k)}$ are the normalized attention coefficients computed by the $k$-th attention mechanism $(a^k)$, and $\\mathbf{W}^{(k)}$ is the corresponding input linear transformation's weight matrix. Note that for this setting, $\\mathbf{h'} \\in \\mathbb{R}^{KF'}$."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "colab": {
     "background_save": true
    },
    "id": "w4j45gTpCeXO"
   },
   "outputs": [],
   "source": [
    "class GAT(MessagePassing):\n",
    "\n",
    "    def __init__(self, in_channels, out_channels, heads = 2,\n",
    "                 negative_slope = 0.2, dropout = 0., **kwargs):\n",
    "        super(GAT, self).__init__(node_dim=0, **kwargs)\n",
    "\n",
    "        self.in_channels = in_channels\n",
    "        self.out_channels = out_channels\n",
    "        self.heads = heads\n",
    "        self.negative_slope = negative_slope\n",
    "        self.dropout = dropout\n",
    "\n",
    "        self.lin_l = None\n",
    "        self.lin_r = None\n",
    "        self.att_l = None\n",
    "        self.att_r = None\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Define the layers needed for the message functions below.\n",
    "        # self.lin_l is the linear transformation that you apply to embeddings \n",
    "        # BEFORE message passing.\n",
    "        # Pay attention to dimensions of the linear layers, since we're using \n",
    "        # multi-head attention.\n",
    "        # Our implementation is ~1 lines, but don't worry if you deviate from this.\n",
    "#         assert (self.out_channels % self.heads == 0), \"Out channel size needs to be div by heads\"\n",
    "        self.lin_l = nn.Linear(self.in_channels, self.heads*self.out_channels)\n",
    "\n",
    "        ############################################################################\n",
    "\n",
    "        self.lin_r = self.lin_l\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Define the attention parameters \\overrightarrow{a_l/r}^T in the above intro.\n",
    "        # You have to deal with multi-head scenarios.\n",
    "        # Use nn.Parameter instead of nn.Linear\n",
    "        # Our implementation is ~2 lines, but don't worry if you deviate from this.\n",
    "        self.att_l = nn.Parameter(torch.randn(heads, self.out_channels)) #Head x C\n",
    "        self.att_r = nn.Parameter(torch.randn(heads, self.out_channels)) #Head x C\n",
    "\n",
    "\n",
    "        ############################################################################\n",
    "\n",
    "        self.reset_parameters()\n",
    "\n",
    "    def reset_parameters(self):\n",
    "        nn.init.xavier_uniform_(self.lin_l.weight)\n",
    "        nn.init.xavier_uniform_(self.lin_r.weight)\n",
    "        nn.init.xavier_uniform_(self.att_l)\n",
    "        nn.init.xavier_uniform_(self.att_r)\n",
    "\n",
    "    def forward(self, x, edge_index, size = None):\n",
    "        \n",
    "        H, C = self.heads, self.out_channels\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Implement message passing, as well as any pre- and post-processing (our update rule).\n",
    "        # 1. First apply linear transformation to node embeddings, and split that \n",
    "        #    into multiple heads. We use the same representations for source and\n",
    "        #    target nodes, but apply different linear weights (W_l and W_r)\n",
    "        # 2. Calculate alpha vectors for central nodes (alpha_l) and neighbor nodes (alpha_r).\n",
    "        # 3. Call propagate function to conduct the message passing. \n",
    "        #    3.1 Remember to pass alpha = (alpha_l, alpha_r) as a parameter.\n",
    "        #    3.2 See there for more information: https://pytorch-geometric.readthedocs.io/en/latest/notes/create_gnn.html\n",
    "        # 4. Transform the output back to the shape of N * d.\n",
    "        # Our implementation is ~5 lines, but don't worry if you deviate from this.\n",
    "        x_l = self.lin_l(x).view(-1, H, C) #N x H x C\n",
    "        x_r = self.lin_r(x).view(-1, H, C) #N x H x C\n",
    "                                                \n",
    "        alpha_l = self.att_l.unsqueeze(0)*x_l #1 x H x C * N x H x C \n",
    "        alpha_r = self.att_r.unsqueeze(0)*x_r #1 x H x C * N x H x C\n",
    "        \n",
    "        out = self.propagate(edge_index, x=(x_l, x_r), alpha=(alpha_l, alpha_r))\n",
    "        out = out.view(-1, H*C)\n",
    "        \n",
    "        ############################################################################\n",
    "\n",
    "        return out\n",
    "\n",
    "\n",
    "    def message(self, x_j, alpha_j, alpha_i, index, ptr, size_i):\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Implement your message function. Putting the attention in message \n",
    "        # instead of in update is a little tricky.\n",
    "        # 1. Calculate the final attention weights using alpha_i and alpha_j,\n",
    "        #    and apply leaky Relu.\n",
    "        # 2. Calculate softmax over the neighbor nodes for all the nodes. Use \n",
    "        #    torch_geometric.utils.softmax instead of the one in Pytorch.\n",
    "        # 3. Apply dropout to attention weights (alpha).\n",
    "        # 4. Multiply embeddings and attention weights. As a sanity check, the output\n",
    "        #    should be of shape E * H * d.\n",
    "        # 5. ptr (LongTensor, optional): If given, computes the softmax based on\n",
    "        #    sorted inputs in CSR representation. You can simply pass it to softmax.\n",
    "        # Our implementation is ~5 lines, but don't worry if you deviate from this.\n",
    "        alpha_ij = F.leaky_relu(alpha_i + alpha_j, negative_slope=self.negative_slope)\n",
    "        \n",
    "        if ptr is not None:\n",
    "            alpha_ij = softmax(alpha_ij, ptr)\n",
    "        else:\n",
    "            alpha_ij = softmax(alpha_ij, index)\n",
    "            \n",
    "        alpha_ij = F.dropout(alpha_ij, p=self.dropout)\n",
    "        \n",
    "        out = x_j * alpha_ij\n",
    "        ############################################################################\n",
    "\n",
    "        return out\n",
    "\n",
    "\n",
    "    def aggregate(self, inputs, index, dim_size = None):\n",
    "\n",
    "        ############################################################################\n",
    "        # TODO: Your code here! \n",
    "        # Implement your aggregate function here.\n",
    "        # See here as how to use torch_scatter.scatter: https://pytorch-scatter.readthedocs.io/en/latest/_modules/torch_scatter/scatter.html\n",
    "        # Pay attention to \"reduce\" parameter is different from that in GraphSage.\n",
    "        # Our implementation is ~1 lines, but don't worry if you deviate from this.\n",
    "        # The axis along which to index number of nodes.\n",
    "        node_dim = self.node_dim\n",
    "        out = torch_scatter.scatter(inputs, index, dim=node_dim, reduce=\"sum\")\n",
    "\n",
    "        ############################################################################\n",
    "    \n",
    "        return out"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "V2dkgSuWCheU"
   },
   "source": [
    "## Building Optimizers\n",
    "\n",
    "This function has been implemented for you. **For grading purposes please use the default Adam optimizer**, but feel free to play with other types of optimizers on your own."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "id": "f_TIQ8NPCjBP"
   },
   "outputs": [],
   "source": [
    "import torch.optim as optim\n",
    "\n",
    "def build_optimizer(args, params):\n",
    "    weight_decay = args.weight_decay\n",
    "    filter_fn = filter(lambda p : p.requires_grad, params)\n",
    "    if args.opt == 'adam':\n",
    "        optimizer = optim.Adam(filter_fn, lr=args.lr, weight_decay=weight_decay)\n",
    "    elif args.opt == 'sgd':\n",
    "        optimizer = optim.SGD(filter_fn, lr=args.lr, momentum=0.95, weight_decay=weight_decay)\n",
    "    elif args.opt == 'rmsprop':\n",
    "        optimizer = optim.RMSprop(filter_fn, lr=args.lr, weight_decay=weight_decay)\n",
    "    elif args.opt == 'adagrad':\n",
    "        optimizer = optim.Adagrad(filter_fn, lr=args.lr, weight_decay=weight_decay)\n",
    "    if args.opt_scheduler == 'none':\n",
    "        return None, optimizer\n",
    "    elif args.opt_scheduler == 'step':\n",
    "        scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=args.opt_decay_step, gamma=args.opt_decay_rate)\n",
    "    elif args.opt_scheduler == 'cos':\n",
    "        scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=args.opt_restart)\n",
    "    return scheduler, optimizer"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "hBYdWFwYCkwY"
   },
   "source": [
    "## Training and Testing\n",
    "\n",
    "Here we provide you with the functions to train and test. **Please do not modify this part for grading purposes.**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "id": "_tZMWRc8CmGg"
   },
   "outputs": [],
   "source": [
    "import time\n",
    "\n",
    "import networkx as nx\n",
    "import numpy as np\n",
    "import torch\n",
    "import torch.optim as optim\n",
    "\n",
    "from torch_geometric.datasets import TUDataset\n",
    "from torch_geometric.datasets import Planetoid\n",
    "from torch_geometric.data import DataLoader\n",
    "\n",
    "import torch_geometric.nn as pyg_nn\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "\n",
    "def train(dataset, args):\n",
    "    \n",
    "    print(\"Node task. test set size:\", np.sum(dataset[0]['train_mask'].numpy()))\n",
    "    test_loader = loader = DataLoader(dataset, batch_size=args.batch_size, shuffle=True)\n",
    "\n",
    "    # build model\n",
    "    model = GNNStack(dataset.num_node_features, args.hidden_dim, dataset.num_classes, \n",
    "                            args)\n",
    "    scheduler, opt = build_optimizer(args, model.parameters())\n",
    "\n",
    "    # train\n",
    "    losses = []\n",
    "    test_accs = []\n",
    "    for epoch in range(args.epochs):\n",
    "        total_loss = 0\n",
    "        model.train()\n",
    "        for batch in loader:\n",
    "            opt.zero_grad()\n",
    "            pred = model(batch)\n",
    "            label = batch.y\n",
    "            pred = pred[batch.train_mask]\n",
    "            label = label[batch.train_mask]\n",
    "            loss = model.loss(pred, label)\n",
    "            loss.backward()\n",
    "            opt.step()\n",
    "            total_loss += loss.item() * batch.num_graphs\n",
    "        total_loss /= len(loader.dataset)\n",
    "        losses.append(total_loss)\n",
    "\n",
    "        if epoch % 10 == 0:\n",
    "            test_acc = test(test_loader, model)\n",
    "            test_accs.append(test_acc)\n",
    "        else:\n",
    "            test_accs.append(test_accs[-1])\n",
    "    return test_accs, losses\n",
    "\n",
    "def test(loader, model, is_validation=True):\n",
    "    model.eval()\n",
    "\n",
    "    correct = 0\n",
    "    for data in loader:\n",
    "        with torch.no_grad():\n",
    "            # max(dim=1) returns values, indices tuple; only need indices\n",
    "            pred = model(data).max(dim=1)[1]\n",
    "            label = data.y\n",
    "\n",
    "        mask = data.val_mask if is_validation else data.test_mask\n",
    "        # node classification: only evaluate on nodes in test set\n",
    "        pred = pred[mask]\n",
    "        label = data.y[mask]\n",
    "            \n",
    "        correct += pred.eq(label).sum().item()\n",
    "\n",
    "    total = 0\n",
    "    for data in loader.dataset:\n",
    "        total += torch.sum(data.val_mask if is_validation else data.test_mask).item()\n",
    "    return correct / total\n",
    "  \n",
    "class objectview(object):\n",
    "    def __init__(self, d):\n",
    "        self.__dict__ = d\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "L7-h7jIsCns4"
   },
   "source": [
    "## Let's Start the Training!\n",
    "\n",
    "We will be working on the CORA dataset on node-level classification.\n",
    "\n",
    "This part is implemented for you. **For grading purposes, please do not modify the default parameters.** However, feel free to play with different configurations just for fun!\n",
    "\n",
    "**Submit your best accuracy and loss on Gradescope.**"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "id": "qe9B45l9Cpz2"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.x\n",
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.tx\n",
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.allx\n",
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.y\n",
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.ty\n",
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.ally\n",
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.graph\n",
      "Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.test.index\n",
      "Processing...\n",
      "Done!\n",
      "Node task. test set size: 140\n",
      "Maximum accuracy: 0.722\n",
      "Minimum loss: 0.11572734266519547\n",
      "Node task. test set size: 140\n",
      "Maximum accuracy: 0.748\n",
      "Minimum loss: 0.02420857548713684\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "def main():\n",
    "    for args in [\n",
    "        {'model_type': 'GraphSage', 'dataset': 'cora', 'num_layers': 2, 'heads': 1, 'batch_size': 32, 'hidden_dim': 32, 'dropout': 0.5, 'epochs': 500, 'opt': 'adam', 'opt_scheduler': 'none', 'opt_restart': 0, 'weight_decay': 5e-3, 'lr': 0.01},\n",
    "    ]:\n",
    "        args = objectview(args)\n",
    "        for model in ['GraphSage', 'GAT']:\n",
    "            args.model_type = model\n",
    "\n",
    "            # Match the dimension.\n",
    "            if model == 'GAT':\n",
    "                args.heads = 2\n",
    "            else:\n",
    "                args.heads = 1\n",
    "\n",
    "            if args.dataset == 'cora':\n",
    "                dataset = Planetoid(root='/tmp/cora', name='Cora')\n",
    "            else:\n",
    "                raise NotImplementedError(\"Unknown dataset\") \n",
    "            test_accs, losses = train(dataset, args) \n",
    "\n",
    "            print(\"Maximum accuracy: {0}\".format(max(test_accs)))\n",
    "            print(\"Minimum loss: {0}\".format(min(losses)))\n",
    "\n",
    "            plt.title(dataset.name)\n",
    "            plt.plot(losses, label=\"training loss\" + \" - \" + args.model_type)\n",
    "            plt.plot(test_accs, label=\"test accuracy\" + \" - \" + args.model_type)\n",
    "        plt.legend()\n",
    "        plt.show()\n",
    "\n",
    "if __name__ == '__main__':\n",
    "    main()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "gHELqjARZ1W5"
   },
   "source": [
    "## Question 1.1: What is the maximum accuracy you could get on test set for GraphSage? (10 points)\n",
    "\n",
    "Submit your answers on Gradescope."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "PlCtBEBLMBkR"
   },
   "source": [
    "## Question 1.2: What is the maximum accuracy you could get on test set for GAT? (10 points)\n",
    "\n",
    "Submit your answers on Gradescope."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Nwwq0nSdmsOL"
   },
   "source": [
    "# 2 DeepSNAP Basics\n",
    "\n",
    "In previous Colabs we used both of graph class (NetworkX) and tensor (PyG) representations of graphs separately. The graph class `nx.Graph` provides rich analysis and manipulation functionalities, such as the clustering coefficient and PageRank. To feed the graph into the model, we need to transform the graph into tensor representations including edge tensor `edge_index` and node attributes tensors `x` and `y`. But only using tensors (as the graphs formatted in PyG `datasets` and `data`) will make many graph manipulations and analysis less efficient and harder. So, in this Colab we will use DeepSNAP which combines both representations and offers a full pipeline for GNN training / validation / testing.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Sf7vUmdNKCjA"
   },
   "source": [
    "In general, [DeepSNAP](https://github.com/snap-stanford/deepsnap) is a Python library to assist efficient deep learning on graphs. DeepSNAP features in its support for flexible graph manipulation, standard pipeline, heterogeneous graphs and simple API.\n",
    "\n",
    "1. DeepSNAP is easy to be used for the sophisticated graph manipulations, such as feature computation, pretraining, subgraph extraction etc. during/before the training.\n",
    "2. In most frameworks, standard pipelines for node, edge, link, graph-level tasks under inductive or transductive settings are left to the user to code. In practice, there are additional design choices involved (such as how to split dataset for link prediction). DeepSNAP provides such a standard pipeline that greatly saves repetitive coding efforts, and enables fair comparision for models.\n",
    "3. Many real-world graphs are heterogeneous graphs. But packages support for heterogeneous graphs, including data storage and flexible message passing, is lacking. DeepSNAP provides an efficient and flexible heterogeneous graph that supports both the node and edge heterogeneity.\n",
    "\n",
    "[DeepSNAP](https://github.com/snap-stanford/deepsnap) is a newly released project and it is still under development. If you find any bugs or have any improvement ideas, feel free to raise issues or create pull requests on the GitHub directly :)\n",
    "\n",
    "In this Colab, we will focus on DeepSNAP graph manipulations and splitting settings."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "20SvvngpQmmQ"
   },
   "source": [
    "## Setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "id": "zfbBVFmAQlwz"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import networkx as nx\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from deepsnap.graph import Graph\n",
    "from deepsnap.batch import Batch\n",
    "from deepsnap.dataset import GraphDataset\n",
    "from torch_geometric.datasets import Planetoid, TUDataset\n",
    "\n",
    "from torch.utils.data import DataLoader\n",
    "\n",
    "def visualize(G, color_map=None, seed=123):\n",
    "    if color_map is None:\n",
    "        color_map = '#c92506'\n",
    "    plt.figure(figsize=(8, 8))\n",
    "    nodes = nx.draw_networkx_nodes(G, pos=nx.spring_layout(G, seed=seed), \\\n",
    "                                 label=None, node_color=color_map, node_shape='o', node_size=150)\n",
    "    edges = nx.draw_networkx_edges(G, pos=nx.spring_layout(G, seed=seed), alpha=0.5)\n",
    "    if color_map is not None:\n",
    "        plt.scatter([],[], c='#c92506', label='Nodes with label 0', edgecolors=\"black\", s=140)\n",
    "        plt.scatter([],[], c='#fcec00', label='Nodes with label 1', edgecolors=\"black\", s=140)\n",
    "        plt.legend(prop={'size': 13}, handletextpad=0)\n",
    "    nodes.set_edgecolor('black')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "ic-o1P3r6hr2"
   },
   "source": [
    "## DeepSNAP Graph\n",
    "\n",
    "The `deepsnap.graph.Graph` class is the core class of DeepSNAP. It not only represents a graph in tensor format but also references to a graph object from graph manipulation package.\n",
    "\n",
    "Currently DeepSNAP supports [NetworkX](https://networkx.org/) and [Snap.py](https://snap.stanford.edu/snappy/doc/index.html) as the back end graph manipulation package.\n",
    "\n",
    "In this Colab, we will use the NetworkX as the back end graph manipulation package."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Ispq_lIoJl_z"
   },
   "source": [
    "Lets first try to convert a simple random NetworkX graph to a DeepSNAP graph."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "id": "zT5qca3x6XpG"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(0, {'node_feature': tensor([0.5151, 0.9839, 0.2097, 0.2312, 0.7975]), 'node_label': tensor(1)})\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Graph(G=[], edge_index=[2, 524], edge_label_index=[2, 524], node_feature=[100, 5], node_label=[100], node_label_index=[100])\n",
      "Node feature (node_feature) has shape torch.Size([100, 5]) and type torch.float32\n",
      "Node label (node_label) has shape torch.Size([100]) and type torch.int64\n",
      "Edge index (edge_index) has shape torch.Size([2, 524]) and type torch.int64\n",
      "The DeepSNAP graph has <class 'networkx.classes.graph.Graph'> as the internal manupulation graph\n"
     ]
    }
   ],
   "source": [
    "num_nodes = 100\n",
    "p = 0.05\n",
    "seed = 100\n",
    "\n",
    "# Generate a networkx random graph\n",
    "G = nx.gnp_random_graph(num_nodes, p, seed=seed)\n",
    "\n",
    "# Generate some random node features and labels\n",
    "node_feature = {node : torch.rand([5, ]) for node in G.nodes()}\n",
    "node_label = {node : torch.randint(0, 2, ()) for node in G.nodes()}\n",
    "\n",
    "# Set the random features and labels to G\n",
    "nx.set_node_attributes(G, node_feature, name='node_feature')\n",
    "nx.set_node_attributes(G, node_label, name='node_label')\n",
    "\n",
    "# Print one node example\n",
    "for node in G.nodes(data=True):\n",
    "    print(node)\n",
    "    break\n",
    "\n",
    "color_map = ['#c92506' if node[1]['node_label'].item() == 0 else '#fcec00' for node in G.nodes(data=True)]\n",
    "\n",
    "# Visualize the graph\n",
    "visualize(G, color_map=color_map)\n",
    "\n",
    "# Transform the networkx graph into the deepsnap graph\n",
    "graph = Graph(G)\n",
    "\n",
    "# Print out the general deepsnap graph information\n",
    "print(graph)\n",
    "\n",
    "# DeepSNAP will convert node attributes to tensors\n",
    "# Notice the type of tensors\n",
    "print(\"Node feature (node_feature) has shape {} and type {}\".format(graph.node_feature.shape, graph.node_feature.dtype))\n",
    "print(\"Node label (node_label) has shape {} and type {}\".format(graph.node_label.shape, graph.node_label.dtype))\n",
    "\n",
    "# DeepSNAP will also generate the edge_index tensor\n",
    "print(\"Edge index (edge_index) has shape {} and type {}\".format(graph.edge_index.shape, graph.edge_index.dtype))\n",
    "\n",
    "# Different from only storing tensors, deepsnap graph also references to the networkx graph\n",
    "# We will discuss why the reference will be helpful later\n",
    "print(\"The DeepSNAP graph has {} as the internal manupulation graph\".format(type(graph.G)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "PNMbc307KOQD"
   },
   "source": [
    "In DeepSNAP we have three levels of attributes. In this example, we have the **node level** attributes including `node_feature` and `node_label`. The other two levels of attributes are graph and edge attributes. The usage is similar to the node level one except that the feature becomes `edge_feature` or `graph_feature` and label becomes `edge_label` or `graph_label` etc."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "G8Xz58_Da0qL"
   },
   "source": [
    "Similar to the NetworkX graph, we can easily get some basic information of the graph through class properties directly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "id": "dLo4zWAoeg6S"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The random graph has 100 nodes\n",
      "The random graph has 262 edges\n"
     ]
    }
   ],
   "source": [
    "# Number of nodes\n",
    "print(\"The random graph has {} nodes\".format(graph.num_nodes))\n",
    "\n",
    "# Number of edges\n",
    "print(\"The random graph has {} edges\".format(graph.num_edges))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Po7IaRmwblI5"
   },
   "source": [
    "DeepSNAP also provides functions that can automatically transform the PyG datasets into a list of DeepSNAP graphs.\n",
    "\n",
    "Here we transform the CORA dataset into a list of DeepSNAP graphs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "id": "ZFkg2kCgcFwR"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Graph(G=[], edge_index=[2, 10556], edge_label_index=[2, 10556], node_feature=[2708, 1433], node_label=[2708], node_label_index=[2708])\n"
     ]
    }
   ],
   "source": [
    "root = './tmp/cora'\n",
    "name = 'Cora'\n",
    "\n",
    "# The Cora dataset\n",
    "pyg_dataset= Planetoid(root, name)\n",
    "\n",
    "# PyG dataset to a list of deepsnap graphs\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "\n",
    "# Get the first deepsnap graph (CORA only has one graph)\n",
    "graph = graphs[0]\n",
    "print(graph)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "NLm5vVYMAP2x"
   },
   "source": [
    "## Question 2.1: What is the number of classes and number of features in the CORA graph? (5 points)\n",
    "\n",
    "Submit your answers on Gradescope."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "id": "8iF_Kyqr_JbY"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cora has 7 classes\n",
      "Cora has 1433 features\n"
     ]
    }
   ],
   "source": [
    "def get_num_node_classes(graph):\n",
    "    # TODO: Implement this function that takes a deepsnap graph object\n",
    "    # and return the number of node classes of that graph.\n",
    "\n",
    "    num_node_classes = 0\n",
    "\n",
    "    ############# Your code here #############\n",
    "    ## (~1 line of code)\n",
    "    ## Note\n",
    "    ## 1. Colab autocomplete functionality might be useful\n",
    "    ## 2. DeepSNAP documentation might be useful https://snap.stanford.edu/deepsnap/modules/graph.html\n",
    "    num_node_classes = graph.num_node_labels\n",
    "\n",
    "    ##########################################\n",
    "\n",
    "    return num_node_classes\n",
    "\n",
    "def get_num_node_features(graph):\n",
    "    # TODO: Implement this function that takes a deepsnap graph object\n",
    "    # and return the number of node features of that graph.\n",
    "\n",
    "    num_node_features = 0\n",
    "\n",
    "    ############# Your code here #############\n",
    "    ## (~1 line of code)\n",
    "    ## Note\n",
    "    ## 1. Colab autocomplete functionality might be useful\n",
    "    ## 2. DeepSNAP documentation might be useful https://snap.stanford.edu/deepsnap/modules/graph.html\n",
    "    num_node_features = graph.num_node_features\n",
    "\n",
    "    ##########################################\n",
    "\n",
    "    return num_node_features\n",
    "\n",
    "num_node_classes = get_num_node_classes(graph)\n",
    "num_node_features = get_num_node_features(graph)\n",
    "print(\"{} has {} classes\".format(name, num_node_classes))\n",
    "print(\"{} has {} features\".format(name, num_node_features))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "rwKbzhHUAckZ"
   },
   "source": [
    "## DeepSNAP Dataset\n",
    "\n",
    "Now, lets talk about DeepSNAP dataset. A `deepsnap.dataset.GraphDataset` contains a list of `deepsnap.graph.Graph` objects. In addition to list of graphs, you can also specify what task the dataset will be used on, such as node level task (`task=node`), edge level task (`task=link_pred`) and graph level task (`task=graph`).\n",
    "\n",
    "It also contains many other useful parameters during initialization and other functinoalities. If you are interested, you can take a look at the [documentation](https://snap.stanford.edu/deepsnap/modules/dataset.html#deepsnap-graphdataset)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "HSidf9E0hn2s"
   },
   "source": [
    "Lets now use COX2 dataset which contains a list of graphs and specify the task to `graph` when we initialize the DeepSNAP dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "id": "l4kqUldyoaS_"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "GraphDataset(467)\n"
     ]
    }
   ],
   "source": [
    "root = './tmp/cox2'\n",
    "name = 'COX2'\n",
    "\n",
    "# Load the dataset through PyG\n",
    "pyg_dataset = TUDataset(root, name)\n",
    "\n",
    "# Convert to a list of deepsnap graphs\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "\n",
    "# Convert list of deepsnap graphs to deepsnap dataset with specified task=graph\n",
    "dataset = GraphDataset(graphs, task='graph')\n",
    "print(dataset)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "7sCV3xJWCddX"
   },
   "source": [
    "## Question 2.2: What is the label of the graph (index 100 in the COX2 dataset)? (5 points)\n",
    "\n",
    "Submit your answers on Gradescope."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "id": "LIis9oTZAfs3"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Graph(G=[], edge_index=[2, 78], edge_label_index=[2, 78], graph_label=[1], node_feature=[37, 35], node_label_index=[37], task=[])\n",
      "Graph with index 100 has label tensor([0])\n"
     ]
    }
   ],
   "source": [
    "def get_graph_class(dataset, idx):\n",
    "    # TODO: Implement this function that takes a deepsnap dataset object,\n",
    "    # the index of the graph in the dataset, and returns the class/label \n",
    "    # of the graph (in integer).\n",
    "\n",
    "    label = -1\n",
    "\n",
    "    ############# Your code here ############\n",
    "    ## (~1 line of code)\n",
    "    ## Note\n",
    "    ## 1. The label refers to the graph-level attribute\n",
    "    label = dataset[idx].graph_label \n",
    "\n",
    "    #########################################\n",
    "\n",
    "    return label\n",
    "\n",
    "graph_0 = dataset[100]\n",
    "print(graph_0)\n",
    "idx = 100\n",
    "label = get_graph_class(dataset, idx)\n",
    "print('Graph with index {} has label {}'.format(idx, label))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "fKhcVeAhCwoY"
   },
   "source": [
    "## Question 2.3: What is the number of edges for the graph (index 200 in the COX2 dataset)? (5 points)\n",
    "\n",
    "Submit your answers on Gradescope."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "id": "f5m2DOfhBtWv"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Graph with index 200 has 49 edges\n"
     ]
    }
   ],
   "source": [
    "def get_graph_num_edges(dataset, idx):\n",
    "    # TODO: Implement this function that takes a deepsnap dataset object,\n",
    "    # the index of the graph in dataset, and returns the number of \n",
    "    # edges in the graph (in integer).\n",
    "\n",
    "    num_edges = 0\n",
    "\n",
    "    ############# Your code here ############\n",
    "    ## (~1 lines of code)\n",
    "    ## Note\n",
    "    ## 1. You can use the class property directly\n",
    "    num_edges = dataset[idx].num_edges\n",
    "\n",
    "    #########################################\n",
    "\n",
    "    return num_edges\n",
    "\n",
    "idx = 200\n",
    "num_edges = get_graph_num_edges(dataset, idx)\n",
    "print('Graph with index {} has {} edges'.format(idx, num_edges))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "AXa7yIG4E0Fp"
   },
   "source": [
    "# 3 DeepSNAP Advanced\n",
    "\n",
    "We have learned the basic use of DeepSNAP graph and dataset :)\n",
    "\n",
    "Lets move on to some more advanced functionalities.\n",
    "\n",
    "In this section we will use DeepSNAP for faeture computation and transductive/inductive splittings."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "Y5fsGBLY8cxa"
   },
   "source": [
    "## Setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "id": "g-jgRLiQ8cSj"
   },
   "outputs": [],
   "source": [
    "import torch\n",
    "import networkx as nx\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from deepsnap.graph import Graph\n",
    "from deepsnap.batch import Batch\n",
    "from deepsnap.dataset import GraphDataset\n",
    "from torch_geometric.datasets import Planetoid, TUDataset\n",
    "\n",
    "from torch.utils.data import DataLoader"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "HnazPGGAJAZN"
   },
   "source": [
    "## Data Split in Graphs\n",
    "\n",
    "Data splitting in graphs can be much harder than that in CV or NLP.\n",
    "\n",
    "In general, the data splitting in graphs can be divided into two settings, **inductive** and **transductive**."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "l9KG_MhqsWBp"
   },
   "source": [
    "## Inductive Split\n",
    "\n",
    "As what we have learned in the lecture, inductive setting will split multiple graphs into each training/valiation and test sets.\n",
    "\n",
    "Here is an example of DeepSNAP inductive splitting for a list of graphs in the graph level task (graph classification etc.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "id": "Gpc6bTm3GF02"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "COX2 train dataset: GraphDataset(373)\n",
      "COX2 validation dataset: GraphDataset(46)\n",
      "COX2 test dataset: GraphDataset(48)\n"
     ]
    }
   ],
   "source": [
    "root = './tmp/cox2'\n",
    "name = 'COX2'\n",
    "\n",
    "pyg_dataset = TUDataset(root, name)\n",
    "\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "\n",
    "# Here we specify the task as graph-level task such as graph classification\n",
    "task = 'graph'\n",
    "dataset = GraphDataset(graphs, task=task)\n",
    "\n",
    "# Specify transductive=False (inductive)\n",
    "dataset_train, dataset_val, dataset_test = dataset.split(transductive=False, split_ratio=[0.8, 0.1, 0.1])\n",
    "\n",
    "print(\"COX2 train dataset: {}\".format(dataset_train))\n",
    "print(\"COX2 validation dataset: {}\".format(dataset_val))\n",
    "print(\"COX2 test dataset: {}\".format(dataset_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "yWKQwa4WsgQp"
   },
   "source": [
    "## Transductive Split\n",
    "\n",
    "In transductive setting, the training /validation / test sets are on the same graph.\n",
    "\n",
    "Here we transductively split the CORA graph in the node level task. \n",
    "\n",
    "(Notice that in DeepSNAP default setting the split is random, but you can also make a fixed split by specifying `fixed_split=True` when loading the dataset from PyG or changing the `node_label_index` directly)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "id": "b5OdxSg4sfyR"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cora train dataset: GraphDataset(1)\n",
      "Cora validation dataset: GraphDataset(1)\n",
      "Cora test dataset: GraphDataset(1)\n",
      "Original Cora has 2708 nodes\n",
      "After the split, Cora has 2166 training nodes\n",
      "After the split, Cora has 270 validation nodes\n",
      "After the split, Cora has 272 test nodes\n"
     ]
    }
   ],
   "source": [
    "root = './tmp/cora'\n",
    "name = 'Cora'\n",
    "\n",
    "pyg_dataset = Planetoid(root, name)\n",
    "\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "\n",
    "# Here we specify the task as node-level task such as node classification\n",
    "task = 'node'\n",
    "\n",
    "dataset = GraphDataset(graphs, task=task)\n",
    "\n",
    "# Specify we want the transductive splitting\n",
    "dataset_train, dataset_val, dataset_test = dataset.split(transductive=True, split_ratio=[0.8, 0.1, 0.1])\n",
    "\n",
    "print(\"Cora train dataset: {}\".format(dataset_train))\n",
    "print(\"Cora validation dataset: {}\".format(dataset_val))\n",
    "print(\"Cora test dataset: {}\".format(dataset_test))\n",
    "\n",
    "print(\"Original Cora has {} nodes\".format(dataset.num_nodes[0]))\n",
    "\n",
    "# The nodes in each set can be find in node_label_index\n",
    "print(\"After the split, Cora has {} training nodes\".format(dataset_train[0].node_label_index.shape[0]))\n",
    "print(\"After the split, Cora has {} validation nodes\".format(dataset_val[0].node_label_index.shape[0]))\n",
    "print(\"After the split, Cora has {} test nodes\".format(dataset_test[0].node_label_index.shape[0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "d7ePKgM00lGE"
   },
   "source": [
    "## Edge Level Split\n",
    "\n",
    "Compared to the node and graph level splitting, edge level splitting is a little bit tricky ;)\n",
    "\n",
    "Usually in edge level splitting, we need to sample negative edges, split positive edges into different datasets, split training edges into message passing edges and supervision edges, and resample the negative edges during the training etc."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "SnzISX5RoiR6"
   },
   "source": [
    "### All Mode\n",
    "\n",
    "Now lets start with a simpler edge level splitting mode, the `edge_train_mode=\"all\"` mode in DeepSNAP."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "id": "_D104xO6137n"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cora train dataset: GraphDataset(1)\n",
      "Cora validation dataset: GraphDataset(1)\n",
      "Cora test dataset: GraphDataset(1)\n"
     ]
    }
   ],
   "source": [
    "root = './tmp/cora'\n",
    "name = 'Cora'\n",
    "\n",
    "pyg_dataset = Planetoid(root, name)\n",
    "\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "\n",
    "# Specify task as link_pred for edge-level task\n",
    "task = 'link_pred'\n",
    "\n",
    "# Specify the train mode, \"all\" mode is default for deepsnap dataset\n",
    "edge_train_mode = \"all\"\n",
    "\n",
    "dataset = GraphDataset(graphs, task=task, edge_train_mode=edge_train_mode)\n",
    "\n",
    "# Transductive link prediction split\n",
    "dataset_train, dataset_val, dataset_test = dataset.split(transductive=True, split_ratio=[0.8, 0.1, 0.1])\n",
    "\n",
    "print(\"Cora train dataset: {}\".format(dataset_train))\n",
    "print(\"Cora validation dataset: {}\".format(dataset_val))\n",
    "print(\"Cora test dataset: {}\".format(dataset_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "GscopwOXC_Y7"
   },
   "source": [
    "In DeepSNAP, the indices of supervision edges are stored in `edge_label_index` tensor and the corresponding edge labels are stored in `edge_label` tensor."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "id": "XJF8fZnA2eLR"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Original Cora graph has 5278 edges\n",
      "Because Cora graph is undirected, the original edge_index has shape torch.Size([2, 10556])\n",
      "The training set has message passing edge index shape torch.Size([2, 8444])\n",
      "The training set has supervision edge index shape torch.Size([2, 16888])\n",
      "The validation set has message passing edge index shape torch.Size([2, 8444])\n",
      "The validation set has supervision edge index shape torch.Size([2, 2108])\n",
      "The test set has message passing edge index shape torch.Size([2, 9498])\n",
      "The test set has supervision edge index shape torch.Size([2, 2116])\n"
     ]
    }
   ],
   "source": [
    "print(\"Original Cora graph has {} edges\".format(dataset[0].num_edges))\n",
    "print(\"Because Cora graph is undirected, the original edge_index has shape {}\".format(dataset[0].edge_index.shape))\n",
    "\n",
    "print(\"The training set has message passing edge index shape {}\".format(dataset_train[0].edge_index.shape))\n",
    "print(\"The training set has supervision edge index shape {}\".format(dataset_train[0].edge_label_index.shape))\n",
    "\n",
    "print(\"The validation set has message passing edge index shape {}\".format(dataset_val[0].edge_index.shape))\n",
    "print(\"The validation set has supervision edge index shape {}\".format(dataset_val[0].edge_label_index.shape))\n",
    "\n",
    "print(\"The test set has message passing edge index shape {}\".format(dataset_test[0].edge_index.shape))\n",
    "print(\"The test set has supervision edge index shape {}\".format(dataset_test[0].edge_label_index.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "s6BX-I_oEKQX"
   },
   "source": [
    "We can see that both training and validation sets have the same message passing edges (`edge_index`) in the `all` mode. Also, in training set, the postive supervision edges (`edge_label_index`) are same with the message passing edges. However, in the test set the message passing edges are the combination of message passing edges from training and validation sets.\n",
    "\n",
    "Notice that the `edge_label` and `edge_label_index` have included the negative edges (default number of negative edges is same with the number of positive edges).\n",
    "\n",
    "Now, lets implement a function that checks whether two edge index tensors are disjoint and explore more edge splitting properties by using that function."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "YOZHDskbAKN6"
   },
   "source": [
    "## Question 3.1 - 3.5: Implement the function that checks whether two edge_index tensors are disjoint. Then answer the True/False questions below. (5 points)\n",
    "\n",
    "Submit your answers on Gradescope."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "id": "wgRYdyPp8EmO"
   },
   "outputs": [],
   "source": [
    "def edge_indices_disjoint(edge_index_1, edge_index_2):\n",
    "    # TODO: Implement this function that takes two edge index tensors,\n",
    "    # and returns whether these two edge index tensors are disjoint.\n",
    "    disjoint = None\n",
    "\n",
    "    ############# Your code here ############\n",
    "    ## (~5 lines of code)\n",
    "    ## Note\n",
    "    ## 1. Here disjoint means that there is no single edge belongs to either edge index tensors\n",
    "    ## 2. You do not need to consider the undirected case. For example, if edge_index_1 contains\n",
    "    ## edge (a, b) and edge_index_2 contains edge (b, a). We will treat them as disjoint in this\n",
    "    ## function.\n",
    "    edge_index_1_np = edge_index_1.T.detach().cpu().numpy()\n",
    "    edge_index_2_np = edge_index_2.T.detach().cpu().numpy() \n",
    "#     print(edge_index_1.shape)\n",
    "    \n",
    "    intersect_np = \\\n",
    "    [x for x in set(tuple(x) for x in edge_index_1_np) & set(tuple(x) for x in edge_index_2_np)]\n",
    "\n",
    "    disjoint =  len(intersect_np) == 0\n",
    "    #########################################\n",
    "\n",
    "    return disjoint"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "id": "EL4ASIDDEIUf"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3.1 Training (supervision) positve and negative edges are disjoint = True\n",
      "3.2 Validation (supervision) positve and negative edges are disjoint = True\n",
      "3.3 Test (supervision) positve and negative edges are disjoint = True\n",
      "3.4 Test (supervision) positve and validation (supervision) positve edges are disjoint = True\n",
      "3.5 Validation (supervision) positve and training (supervision) positve edges are disjoint = True\n"
     ]
    }
   ],
   "source": [
    "num_train_edges = dataset_train[0].edge_label_index.shape[1] // 2\n",
    "train_pos_edge_index = dataset_train[0].edge_label_index[:, :num_train_edges]\n",
    "train_neg_edge_index = dataset_train[0].edge_label_index[:, num_train_edges:]\n",
    "print(\"3.1 Training (supervision) positve and negative edges are disjoint = {}\"\\\n",
    "        .format(edge_indices_disjoint(train_pos_edge_index, train_neg_edge_index)))\n",
    "\n",
    "num_val_edges = dataset_val[0].edge_label_index.shape[1] // 2\n",
    "val_pos_edge_index = dataset_val[0].edge_label_index[:, :num_val_edges]\n",
    "val_neg_edge_index = dataset_val[0].edge_label_index[:, num_val_edges:]\n",
    "print(\"3.2 Validation (supervision) positve and negative edges are disjoint = {}\"\\\n",
    "        .format(edge_indices_disjoint(val_pos_edge_index, val_neg_edge_index)))\n",
    "\n",
    "num_test_edges = dataset_test[0].edge_label_index.shape[1] // 2\n",
    "test_pos_edge_index = dataset_test[0].edge_label_index[:, :num_test_edges]\n",
    "test_neg_edge_index = dataset_test[0].edge_label_index[:, num_test_edges:]\n",
    "print(\"3.3 Test (supervision) positve and negative edges are disjoint = {}\"\\\n",
    "        .format(edge_indices_disjoint(test_pos_edge_index, test_neg_edge_index)))\n",
    "\n",
    "print(\"3.4 Test (supervision) positve and validation (supervision) positve edges are disjoint = {}\"\\\n",
    "        .format(edge_indices_disjoint(test_pos_edge_index, val_pos_edge_index)))\n",
    "print(\"3.5 Validation (supervision) positve and training (supervision) positve edges are disjoint = {}\"\\\n",
    "        .format(edge_indices_disjoint(val_pos_edge_index, train_pos_edge_index)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "8jLoVN5ZBTuA"
   },
   "source": [
    "### Disjoint Mode\n",
    "\n",
    "Now lets look at a relatively more complex transductive edge split setting, which is the `edge_train_mode=\"disjoint\"` mode in DeepSNAP (also the transductive link prediction splitting talked in the lecture)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "id": "3Rqzfb-0BTBm"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The edge index of original graph has shape: torch.Size([2, 10556])\n",
      "The edge index of training message edges has shape: torch.Size([2, 6754])\n",
      "The edge index of training supervision edges has shape: torch.Size([2, 3380])\n",
      "The edge index of validation message edges has shape: torch.Size([2, 8444])\n",
      "The edge index of validation supervision edges has shape: torch.Size([2, 2108])\n",
      "The edge index of test message edges has shape: torch.Size([2, 9498])\n",
      "The edge index of test supervision edges has shape: torch.Size([2, 2116])\n"
     ]
    }
   ],
   "source": [
    "edge_train_mode = \"disjoint\"\n",
    "\n",
    "dataset = GraphDataset(graphs, task='link_pred', edge_train_mode=edge_train_mode)\n",
    "orig_edge_index = dataset[0].edge_index\n",
    "dataset_train, dataset_val, dataset_test = dataset.split(\n",
    "    transductive=True, split_ratio=[0.8, 0.1, 0.1])\n",
    "\n",
    "train_message_edge_index = dataset_train[0].edge_index\n",
    "train_sup_edge_index = dataset_train[0].edge_label_index\n",
    "val_sup_edge_index = dataset_val[0].edge_label_index\n",
    "test_sup_edge_index = dataset_test[0].edge_label_index\n",
    "\n",
    "print(\"The edge index of original graph has shape: {}\".format(orig_edge_index.shape))\n",
    "print(\"The edge index of training message edges has shape: {}\".format(train_message_edge_index.shape))\n",
    "print(\"The edge index of training supervision edges has shape: {}\".format(train_sup_edge_index.shape))\n",
    "print(\"The edge index of validation message edges has shape: {}\".format(dataset_val[0].edge_index.shape))\n",
    "print(\"The edge index of validation supervision edges has shape: {}\".format(val_sup_edge_index.shape))\n",
    "print(\"The edge index of test message edges has shape: {}\".format(dataset_test[0].edge_index.shape))\n",
    "print(\"The edge index of test supervision edges has shape: {}\".format(test_sup_edge_index.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "TUkBhiJNciol"
   },
   "source": [
    "You can see that the training / validation message passing edges and training supervision edges are splitted differently in those two modes!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "2WKfRjqAJHtK"
   },
   "source": [
    "### Resample Negative Edges\n",
    "\n",
    "During each training iteration, we usually need to resample the negative edges.\n",
    "\n",
    "Below we print the training and validation sets negative edges in two training iterations.\n",
    "\n",
    "You should find that the negative edges in training set will be resampled."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "id": "AMEbnx63JHWj"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "First iteration training negative edges:\n",
      "tensor([[ 141, 2360, 1231,  ..., 2455, 1427, 1147],\n",
      "        [1866, 1002, 2184,  ..., 1252, 2705, 1259]])\n",
      "Second iteration training negative edges:\n",
      "tensor([[1711, 2118, 2070,  ..., 1100,  613,  487],\n",
      "        [  70,  546,  477,  ..., 1403,  230,  401]])\n",
      "First iteration validation negative edges:\n",
      "tensor([[1459, 1770, 2162,  ..., 1409, 1133, 2258],\n",
      "        [ 266,  764, 1344,  ..., 1337,  663,  799]])\n",
      "Second iteration validation negative edges:\n",
      "tensor([[1459, 1770, 2162,  ..., 1409, 1133, 2258],\n",
      "        [ 266,  764, 1344,  ..., 1337,  663,  799]])\n"
     ]
    }
   ],
   "source": [
    "dataset = GraphDataset(graphs, task='link_pred', edge_train_mode=\"disjoint\")\n",
    "datasets = {}\n",
    "follow_batch = []\n",
    "datasets['train'], datasets['val'], datasets['test'] = dataset.split(\n",
    "    transductive=True, split_ratio=[0.8, 0.1, 0.1])\n",
    "dataloaders = {\n",
    "  split: DataLoader(\n",
    "    ds, collate_fn=Batch.collate(follow_batch),\n",
    "    batch_size=1, shuffle=(split=='train')\n",
    "  )\n",
    "  for split, ds in datasets.items()\n",
    "}\n",
    "neg_edges_1 = None\n",
    "for batch in dataloaders['train']:\n",
    "    num_edges = batch.edge_label_index.shape[1] // 2\n",
    "    neg_edges_1 = batch.edge_label_index[:, num_edges:]\n",
    "    print(\"First iteration training negative edges:\")\n",
    "    print(neg_edges_1)\n",
    "    break\n",
    "neg_edges_2 = None\n",
    "for batch in dataloaders['train']:\n",
    "    num_edges = batch.edge_label_index.shape[1] // 2\n",
    "    neg_edges_2 = batch.edge_label_index[:, num_edges:]\n",
    "    print(\"Second iteration training negative edges:\")\n",
    "    print(neg_edges_2)\n",
    "    break\n",
    "\n",
    "neg_edges_1 = None\n",
    "for batch in dataloaders['val']:\n",
    "    num_edges = batch.edge_label_index.shape[1] // 2\n",
    "    neg_edges_1 = batch.edge_label_index[:, num_edges:]\n",
    "    print(\"First iteration validation negative edges:\")\n",
    "    print(neg_edges_1)\n",
    "    break\n",
    "neg_edges_2 = None\n",
    "for batch in dataloaders['val']:\n",
    "    num_edges = batch.edge_label_index.shape[1] // 2\n",
    "    neg_edges_2 = batch.edge_label_index[:, num_edges:]\n",
    "    print(\"Second iteration validation negative edges:\")\n",
    "    print(neg_edges_2)\n",
    "    break"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "DEzqh7wEdrh0"
   },
   "source": [
    "If you are interested in more graph splitting settings, please refer to the DeepSNAP dataset [documentation](https://snap.stanford.edu/deepsnap/modules/dataset.html)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "XkrYyeSUI_9_"
   },
   "source": [
    "## Graph Transformation and Feature Computation\n",
    "\n",
    "The other DeepSNAP core functionality is graph transformation / feature computation.\n",
    "\n",
    "In DeepSNAP, we divide graph transformation / feature computation into two different types. One is the transformation before training (transform the whole dataset before training directly) and another one is the transformation during training (transform batches of graphs).\n",
    "\n",
    "Here is an example that uses NetworkX back end to calculate the PageRank value and update the value to tensors before the training (transform the dataset)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "id": "gnAVbZINLZ4I"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of features before transformation: 35\n",
      "Number of features after transformation: 36\n"
     ]
    }
   ],
   "source": [
    "def pagerank_transform_fn(graph):\n",
    "\n",
    "    # Get the referenced networkx graph\n",
    "    G = graph.G\n",
    "\n",
    "    # Calculate the pagerank by using networkx\n",
    "    pr = nx.pagerank(G)\n",
    "\n",
    "    # Transform the pagerank values to tensor\n",
    "    pr_feature = torch.tensor([pr[node] for node in range(graph.num_nodes)], dtype=torch.float32)\n",
    "    pr_feature = pr_feature.view(graph.num_nodes, 1)\n",
    "\n",
    "    # Concat the pagerank values to the node feature\n",
    "    graph.node_feature = torch.cat([graph.node_feature, pr_feature], dim=-1)\n",
    "\n",
    "root = './tmp/cox2'\n",
    "name = 'COX2'\n",
    "pyg_dataset = TUDataset(root, name)\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "dataset = GraphDataset(graphs, task='graph')\n",
    "print(\"Number of features before transformation: {}\".format(dataset.num_node_features))\n",
    "dataset.apply_transform(pagerank_transform_fn, update_tensor=False)\n",
    "print(\"Number of features after transformation: {}\".format(dataset.num_node_features))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "LHByE87SQkUw"
   },
   "source": [
    "## Question 3.6: Implement the transformation below and report the clustering coefficient of the node (index 3) of the graph (index 406) in the COX2 dataset. Rounded the answer to two decimal places. (5 points)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "id": "RNEjfOZRNjYb"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The node has clustering coefficient: 0.17\n"
     ]
    }
   ],
   "source": [
    "def cluster_transform_fn(graph):\n",
    "    # TODO: Implement this function that takes an deepsnap graph object,\n",
    "    # transform the graph by adding nodes clustering coefficient into the \n",
    "    # graph.node_feature\n",
    "\n",
    "    ############# Your code here ############\n",
    "    ## (~5 lines of code)\n",
    "    ## Note\n",
    "    ## 1. Compute the clustering coefficient value for each node and\n",
    "    ## concat them to the last dimension of graph.node_feature\n",
    "    \n",
    "    # Get the referenced networkx graph\n",
    "    G = graph.G\n",
    "\n",
    "    # Calculate the cluster coef by using networkx\n",
    "    pr = nx.algorithms.cluster.clustering(G)\n",
    "\n",
    "    # Transform the pagerank values to tensor\n",
    "    pr_feature = torch.tensor([pr[node] for node in range(graph.num_nodes)], dtype=torch.float32)\n",
    "    pr_feature = pr_feature.view(graph.num_nodes, 1)\n",
    "\n",
    "    # Concat the pagerank values to the node feature\n",
    "    graph.node_feature = torch.cat([graph.node_feature, pr_feature], dim=-1)\n",
    "\n",
    "\n",
    "    #########################################\n",
    "\n",
    "root = './cox2'\n",
    "name = 'COX2'\n",
    "pyg_dataset = TUDataset(root, name)\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "dataset = GraphDataset(graphs, task='graph')\n",
    "\n",
    "# Transform the dataset\n",
    "dataset.apply_transform(cluster_transform_fn, update_tensor=False)\n",
    "\n",
    "node_idx = 3\n",
    "graph_idx = 406\n",
    "node_feature = dataset[graph_idx].node_feature\n",
    "\n",
    "print(\"The node has clustering coefficient: {}\".format(round(node_feature[node_idx][-1].item(), 2)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "4P5Ig7XaPYzp"
   },
   "source": [
    "Apart from transforming the dataset, DeepSNAP can also transform the graph (usually the `deepsnap.batch.Batch`) during each training iteration.\n",
    "\n",
    "Also, DeepSNAP supports the synchronization of the transformation between the referenced graph objects and tensor representations. For example, you can just update the NetworkX graph object in the transform function, and by specifying `update_tensor=True` the internal tensor representations will be automatically updated.\n",
    "\n",
    "For more information, please refer to the DeepSNAP [documentation](https://snap.stanford.edu/deepsnap/)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "s-YLYMLFQYqp"
   },
   "source": [
    "# 4 Edge Level Prediction\n",
    "\n",
    "From last section, we know how DeepSNAP transductive split the edges in the link prediction task.\n",
    "\n",
    "Now lets use DeepSNAP and PyG together to implement a edge level prediction (link prediction) model!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "id": "hrKCNtvERypQ"
   },
   "outputs": [],
   "source": [
    "import copy\n",
    "import torch\n",
    "import numpy as np\n",
    "import networkx as nx\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from deepsnap.graph import Graph\n",
    "from deepsnap.batch import Batch\n",
    "from deepsnap.dataset import GraphDataset\n",
    "from torch_geometric.datasets import Planetoid, TUDataset\n",
    "\n",
    "from torch.utils.data import DataLoader\n",
    "\n",
    "import torch.nn.functional as F\n",
    "from torch_geometric.nn import SAGEConv\n",
    "\n",
    "class LinkPredModel(torch.nn.Module):\n",
    "    def __init__(self, input_dim, hidden_dim, num_classes, dropout=0.2):\n",
    "        super(LinkPredModel, self).__init__()\n",
    "\n",
    "        self.conv1 = SAGEConv(input_dim, hidden_dim)\n",
    "        self.conv2 = SAGEConv(hidden_dim, num_classes)\n",
    "\n",
    "        self.loss_fn = None\n",
    "\n",
    "        ############# Your code here #############\n",
    "        ## (~1 line of code)\n",
    "        ## Note\n",
    "        ## 1. Initialize the loss function to BCEWithLogitsLoss\n",
    "        self.loss_fn = nn.BCEWithLogitsLoss()\n",
    "\n",
    "        ##########################################\n",
    "\n",
    "        self.dropout = dropout\n",
    "\n",
    "    def reset_parameters(self):\n",
    "        self.conv1.reset_parameters()\n",
    "        self.conv2.reset_parameters()\n",
    "\n",
    "    def forward(self, batch):\n",
    "        node_feature, edge_index, edge_label_index = batch.node_feature, batch.edge_index, batch.edge_label_index\n",
    "        \n",
    "        ############# Your code here #############\n",
    "        ## (~6 line of code)\n",
    "        ## Note\n",
    "        ## 1. Feed the node feature into the first conv layer\n",
    "        ## 2. Add a ReLU after the first conv layer\n",
    "        ## 3. Add dropout after the ReLU (with probability self.dropout)\n",
    "        ## 4. Feed the output to the second conv layer\n",
    "        ## 5. Select the embeddings of the source nodes and destination nodes\n",
    "        ## by using the edge_label_index and compute the similarity of each pair\n",
    "        ## by dot product\n",
    "        x = F.relu(self.conv1(node_feature, edge_index))\n",
    "        x = F.dropout(x, p=self.dropout)\n",
    "        x = self.conv2(x, edge_index)\n",
    "        \n",
    "        x_src = x[edge_label_index[0]]\n",
    "        x_tgt = x[edge_label_index[1]]\n",
    "        \n",
    "        x_sim = x_src * x_tgt\n",
    "        pred = torch.sum(x_sim, dim=-1)\n",
    "        \n",
    "        ##########################################\n",
    "\n",
    "        return pred\n",
    "    \n",
    "    def loss(self, pred, link_label):\n",
    "        return self.loss_fn(pred, link_label)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0., 0., 1.])"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_pred = np.array([0.2, 0.1, 0.9])\n",
    "test_pred.round()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "id": "uuKbGFOu1Ka8"
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import *\n",
    "\n",
    "def train(model, dataloaders, optimizer, args):\n",
    "    val_max = 0\n",
    "    best_model = model\n",
    "\n",
    "    for epoch in range(1, args[\"epochs\"]):\n",
    "        for i, batch in enumerate(dataloaders['train']):\n",
    "            \n",
    "            batch.to(args[\"device\"])\n",
    "\n",
    "            ############# Your code here #############\n",
    "            ## (~6 lines of code)\n",
    "            ## Note\n",
    "            ## 1. Zero grad the optimizer\n",
    "            ## 2. Compute loss and backpropagate\n",
    "            ## 3. Update the model parameters\n",
    "            optimizer.zero_grad()\n",
    "            pred = model(batch)\n",
    "            loss = model.loss(pred, batch.edge_label.type_as(pred))\n",
    "            \n",
    "            loss.backward()\n",
    "            optimizer.step()\n",
    "\n",
    "            ##########################################\n",
    "\n",
    "            log = 'Epoch: {:03d}, Train: {:.4f}, Val: {:.4f}, Test: {:.4f}, Loss: {}'\n",
    "            score_train = test(model, dataloaders['train'], args)\n",
    "            score_val = test(model, dataloaders['val'], args)\n",
    "            score_test = test(model, dataloaders['test'], args)\n",
    "\n",
    "            print(log.format(epoch, score_train, score_val, score_test, loss.item()))\n",
    "            if val_max < score_val:\n",
    "                val_max = score_val\n",
    "                best_model = copy.deepcopy(model)\n",
    "    return best_model\n",
    "\n",
    "def test(model, dataloader, args):\n",
    "    model.eval()\n",
    "\n",
    "    score = 0\n",
    "\n",
    "    ############# Your code here #############\n",
    "    ## (~5 lines of code)\n",
    "    ## Note\n",
    "    ## 1. Loop through batches in the dataloader\n",
    "    ## 2. Feed the batch to the model\n",
    "    ## 3. Feed the model output to sigmoid\n",
    "    ## 4. Compute the ROC-AUC score by using sklearn roc_auc_score function\n",
    "    ## 5. Edge labels are stored in batch.edge_label\n",
    "    for batch in dataloader:\n",
    "        batch.to(args[\"device\"])\n",
    "        \n",
    "        pred = model(batch)\n",
    "        pred = torch.sigmoid(pred).cpu().detach().numpy()\n",
    "        label = batch.edge_label.cpu().detach().numpy()\n",
    "        score += roc_auc_score(label, pred)\n",
    "#         score += accuracy_score(label, pred.round())\n",
    "        \n",
    "    score /= len(dataloader)\n",
    "    ##########################################\n",
    " \n",
    "    return score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "id": "JTKWYX1b33V3"
   },
   "outputs": [],
   "source": [
    "# Please don't change any parameters\n",
    "args = {\n",
    "    \"device\" : 'cuda' if torch.cuda.is_available() else 'cpu',\n",
    "    \"hidden_dim\" : 128,\n",
    "    \"epochs\" : 200,\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "id": "Klw_xYnE27xQ",
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 001, Train: 0.5463, Val: 0.5016, Test: 0.5421, Loss: 0.6930463314056396\n",
      "Epoch: 002, Train: 0.5435, Val: 0.4930, Test: 0.5311, Loss: 0.6930522918701172\n",
      "Epoch: 003, Train: 0.5405, Val: 0.5195, Test: 0.5336, Loss: 0.6930674314498901\n",
      "Epoch: 004, Train: 0.5721, Val: 0.5181, Test: 0.5617, Loss: 0.6930385231971741\n",
      "Epoch: 005, Train: 0.5591, Val: 0.5270, Test: 0.5504, Loss: 0.6930369734764099\n",
      "Epoch: 006, Train: 0.5551, Val: 0.5184, Test: 0.5526, Loss: 0.6930408477783203\n",
      "Epoch: 007, Train: 0.5425, Val: 0.5257, Test: 0.5101, Loss: 0.6930454969406128\n",
      "Epoch: 008, Train: 0.5530, Val: 0.5100, Test: 0.5507, Loss: 0.6929976940155029\n",
      "Epoch: 009, Train: 0.5531, Val: 0.5268, Test: 0.5315, Loss: 0.6930065155029297\n",
      "Epoch: 010, Train: 0.5704, Val: 0.5285, Test: 0.5511, Loss: 0.6930297017097473\n",
      "Epoch: 011, Train: 0.5637, Val: 0.5064, Test: 0.5390, Loss: 0.6930392384529114\n",
      "Epoch: 012, Train: 0.5533, Val: 0.5255, Test: 0.5475, Loss: 0.6930128931999207\n",
      "Epoch: 013, Train: 0.5638, Val: 0.5284, Test: 0.5453, Loss: 0.6930238604545593\n",
      "Epoch: 014, Train: 0.5566, Val: 0.5506, Test: 0.5367, Loss: 0.6929939985275269\n",
      "Epoch: 015, Train: 0.5785, Val: 0.5360, Test: 0.5591, Loss: 0.6929974555969238\n",
      "Epoch: 016, Train: 0.5711, Val: 0.5034, Test: 0.5497, Loss: 0.6929904222488403\n",
      "Epoch: 017, Train: 0.5595, Val: 0.5296, Test: 0.5452, Loss: 0.6929751038551331\n",
      "Epoch: 018, Train: 0.5806, Val: 0.4988, Test: 0.5490, Loss: 0.6929959058761597\n",
      "Epoch: 019, Train: 0.5664, Val: 0.5051, Test: 0.5572, Loss: 0.6929736733436584\n",
      "Epoch: 020, Train: 0.5728, Val: 0.5270, Test: 0.5561, Loss: 0.6929952502250671\n",
      "Epoch: 021, Train: 0.5717, Val: 0.5199, Test: 0.5665, Loss: 0.6929189562797546\n",
      "Epoch: 022, Train: 0.5635, Val: 0.5374, Test: 0.5567, Loss: 0.6929086446762085\n",
      "Epoch: 023, Train: 0.5673, Val: 0.5452, Test: 0.5547, Loss: 0.6929723620414734\n",
      "Epoch: 024, Train: 0.5725, Val: 0.5147, Test: 0.5563, Loss: 0.6929633021354675\n",
      "Epoch: 025, Train: 0.5988, Val: 0.5482, Test: 0.5518, Loss: 0.6929066181182861\n",
      "Epoch: 026, Train: 0.5737, Val: 0.5247, Test: 0.5399, Loss: 0.6929170489311218\n",
      "Epoch: 027, Train: 0.5704, Val: 0.5064, Test: 0.5584, Loss: 0.6928554773330688\n",
      "Epoch: 028, Train: 0.5835, Val: 0.5080, Test: 0.5630, Loss: 0.6928803324699402\n",
      "Epoch: 029, Train: 0.5783, Val: 0.5250, Test: 0.5526, Loss: 0.6928356289863586\n",
      "Epoch: 030, Train: 0.5924, Val: 0.5410, Test: 0.5576, Loss: 0.6928685307502747\n",
      "Epoch: 031, Train: 0.5760, Val: 0.5169, Test: 0.5581, Loss: 0.6928555965423584\n",
      "Epoch: 032, Train: 0.5768, Val: 0.5237, Test: 0.5532, Loss: 0.6928229928016663\n",
      "Epoch: 033, Train: 0.5819, Val: 0.5384, Test: 0.5447, Loss: 0.6927967667579651\n",
      "Epoch: 034, Train: 0.5831, Val: 0.5442, Test: 0.5535, Loss: 0.6927506923675537\n",
      "Epoch: 035, Train: 0.5765, Val: 0.5234, Test: 0.5512, Loss: 0.6926714181900024\n",
      "Epoch: 036, Train: 0.5843, Val: 0.5393, Test: 0.5643, Loss: 0.6927794814109802\n",
      "Epoch: 037, Train: 0.5918, Val: 0.5323, Test: 0.5458, Loss: 0.692727267742157\n",
      "Epoch: 038, Train: 0.5924, Val: 0.5350, Test: 0.5458, Loss: 0.6926540732383728\n",
      "Epoch: 039, Train: 0.5830, Val: 0.5234, Test: 0.5319, Loss: 0.6925792098045349\n",
      "Epoch: 040, Train: 0.5937, Val: 0.5417, Test: 0.5586, Loss: 0.6925698518753052\n",
      "Epoch: 041, Train: 0.5991, Val: 0.5203, Test: 0.5384, Loss: 0.6925806999206543\n",
      "Epoch: 042, Train: 0.5812, Val: 0.5276, Test: 0.5505, Loss: 0.692432701587677\n",
      "Epoch: 043, Train: 0.5925, Val: 0.5247, Test: 0.5490, Loss: 0.6924381852149963\n",
      "Epoch: 044, Train: 0.5843, Val: 0.5395, Test: 0.5379, Loss: 0.6925178170204163\n",
      "Epoch: 045, Train: 0.5875, Val: 0.5275, Test: 0.5519, Loss: 0.6923069953918457\n",
      "Epoch: 046, Train: 0.5867, Val: 0.5142, Test: 0.5651, Loss: 0.692264974117279\n",
      "Epoch: 047, Train: 0.6031, Val: 0.5280, Test: 0.5499, Loss: 0.6922422647476196\n",
      "Epoch: 048, Train: 0.5945, Val: 0.5427, Test: 0.5450, Loss: 0.6923220753669739\n",
      "Epoch: 049, Train: 0.6119, Val: 0.5396, Test: 0.5492, Loss: 0.6921339631080627\n",
      "Epoch: 050, Train: 0.5993, Val: 0.5457, Test: 0.5449, Loss: 0.6920750737190247\n",
      "Epoch: 051, Train: 0.6057, Val: 0.5398, Test: 0.5492, Loss: 0.6920371651649475\n",
      "Epoch: 052, Train: 0.6115, Val: 0.5479, Test: 0.5533, Loss: 0.6916666030883789\n",
      "Epoch: 053, Train: 0.6162, Val: 0.5223, Test: 0.5475, Loss: 0.6917535662651062\n",
      "Epoch: 054, Train: 0.6093, Val: 0.5556, Test: 0.5438, Loss: 0.691766619682312\n",
      "Epoch: 055, Train: 0.5983, Val: 0.5448, Test: 0.5510, Loss: 0.6915207505226135\n",
      "Epoch: 056, Train: 0.5934, Val: 0.5452, Test: 0.5451, Loss: 0.6913994550704956\n",
      "Epoch: 057, Train: 0.5976, Val: 0.5130, Test: 0.5603, Loss: 0.69135582447052\n",
      "Epoch: 058, Train: 0.5928, Val: 0.5219, Test: 0.5585, Loss: 0.6911681294441223\n",
      "Epoch: 059, Train: 0.5979, Val: 0.5435, Test: 0.5476, Loss: 0.6910039782524109\n",
      "Epoch: 060, Train: 0.6130, Val: 0.5458, Test: 0.5565, Loss: 0.690961480140686\n",
      "Epoch: 061, Train: 0.6080, Val: 0.5317, Test: 0.5531, Loss: 0.6907222867012024\n",
      "Epoch: 062, Train: 0.6243, Val: 0.5362, Test: 0.5552, Loss: 0.6907998323440552\n",
      "Epoch: 063, Train: 0.6111, Val: 0.5549, Test: 0.5582, Loss: 0.6905665993690491\n",
      "Epoch: 064, Train: 0.6066, Val: 0.5324, Test: 0.5527, Loss: 0.6901450753211975\n",
      "Epoch: 065, Train: 0.5984, Val: 0.5492, Test: 0.5584, Loss: 0.6900092363357544\n",
      "Epoch: 066, Train: 0.6144, Val: 0.5324, Test: 0.5503, Loss: 0.6897639036178589\n",
      "Epoch: 067, Train: 0.6103, Val: 0.5384, Test: 0.5511, Loss: 0.6893777251243591\n",
      "Epoch: 068, Train: 0.6260, Val: 0.5473, Test: 0.5566, Loss: 0.6893033385276794\n",
      "Epoch: 069, Train: 0.6078, Val: 0.5447, Test: 0.5557, Loss: 0.6889249086380005\n",
      "Epoch: 070, Train: 0.6311, Val: 0.5443, Test: 0.5589, Loss: 0.689019501209259\n",
      "Epoch: 071, Train: 0.6328, Val: 0.5548, Test: 0.5590, Loss: 0.688266396522522\n",
      "Epoch: 072, Train: 0.6241, Val: 0.5491, Test: 0.5650, Loss: 0.6884793043136597\n",
      "Epoch: 073, Train: 0.6252, Val: 0.5489, Test: 0.5593, Loss: 0.6881080865859985\n",
      "Epoch: 074, Train: 0.6194, Val: 0.5540, Test: 0.5629, Loss: 0.6867387890815735\n",
      "Epoch: 075, Train: 0.6271, Val: 0.5460, Test: 0.5582, Loss: 0.6873410940170288\n",
      "Epoch: 076, Train: 0.6198, Val: 0.5533, Test: 0.5665, Loss: 0.6857807040214539\n",
      "Epoch: 077, Train: 0.6324, Val: 0.5509, Test: 0.5723, Loss: 0.6860828995704651\n",
      "Epoch: 078, Train: 0.6284, Val: 0.5638, Test: 0.5650, Loss: 0.685697078704834\n",
      "Epoch: 079, Train: 0.6254, Val: 0.5602, Test: 0.5735, Loss: 0.6848316192626953\n",
      "Epoch: 080, Train: 0.6440, Val: 0.5619, Test: 0.5727, Loss: 0.6842788457870483\n",
      "Epoch: 081, Train: 0.6367, Val: 0.5527, Test: 0.5714, Loss: 0.6841903924942017\n",
      "Epoch: 082, Train: 0.6318, Val: 0.5592, Test: 0.5779, Loss: 0.6837276816368103\n",
      "Epoch: 083, Train: 0.6300, Val: 0.5767, Test: 0.5793, Loss: 0.6829962730407715\n",
      "Epoch: 084, Train: 0.6367, Val: 0.5722, Test: 0.5708, Loss: 0.6818441152572632\n",
      "Epoch: 085, Train: 0.6392, Val: 0.5652, Test: 0.5750, Loss: 0.6809617877006531\n",
      "Epoch: 086, Train: 0.6475, Val: 0.5716, Test: 0.5810, Loss: 0.6810272932052612\n",
      "Epoch: 087, Train: 0.6387, Val: 0.5858, Test: 0.5855, Loss: 0.6778398752212524\n",
      "Epoch: 088, Train: 0.6589, Val: 0.5901, Test: 0.5870, Loss: 0.6778855323791504\n",
      "Epoch: 089, Train: 0.6514, Val: 0.5890, Test: 0.5801, Loss: 0.6764360666275024\n",
      "Epoch: 090, Train: 0.6418, Val: 0.5899, Test: 0.5928, Loss: 0.6762486100196838\n",
      "Epoch: 091, Train: 0.6483, Val: 0.5928, Test: 0.5900, Loss: 0.6730237007141113\n",
      "Epoch: 092, Train: 0.6707, Val: 0.5893, Test: 0.5933, Loss: 0.6747995615005493\n",
      "Epoch: 093, Train: 0.6574, Val: 0.6015, Test: 0.5996, Loss: 0.6716553568840027\n",
      "Epoch: 094, Train: 0.6605, Val: 0.6148, Test: 0.6115, Loss: 0.6683481931686401\n",
      "Epoch: 095, Train: 0.6705, Val: 0.6127, Test: 0.6137, Loss: 0.6716288328170776\n",
      "Epoch: 096, Train: 0.6823, Val: 0.6248, Test: 0.6202, Loss: 0.6669378280639648\n",
      "Epoch: 097, Train: 0.6745, Val: 0.6267, Test: 0.6259, Loss: 0.6642219424247742\n",
      "Epoch: 098, Train: 0.6837, Val: 0.6337, Test: 0.6323, Loss: 0.6665745973587036\n",
      "Epoch: 099, Train: 0.6890, Val: 0.6401, Test: 0.6271, Loss: 0.660476803779602\n",
      "Epoch: 100, Train: 0.6916, Val: 0.6604, Test: 0.6462, Loss: 0.659335196018219\n",
      "Epoch: 101, Train: 0.7030, Val: 0.6590, Test: 0.6506, Loss: 0.6556150913238525\n",
      "Epoch: 102, Train: 0.6919, Val: 0.6570, Test: 0.6554, Loss: 0.6550018787384033\n",
      "Epoch: 103, Train: 0.7181, Val: 0.6779, Test: 0.6547, Loss: 0.6539687514305115\n",
      "Epoch: 104, Train: 0.7131, Val: 0.6804, Test: 0.6578, Loss: 0.6469799876213074\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 105, Train: 0.7133, Val: 0.6890, Test: 0.6647, Loss: 0.6486737728118896\n",
      "Epoch: 106, Train: 0.7335, Val: 0.6819, Test: 0.6692, Loss: 0.6451020240783691\n",
      "Epoch: 107, Train: 0.7228, Val: 0.6919, Test: 0.6717, Loss: 0.6419759392738342\n",
      "Epoch: 108, Train: 0.7246, Val: 0.6924, Test: 0.6711, Loss: 0.6404592990875244\n",
      "Epoch: 109, Train: 0.7365, Val: 0.6939, Test: 0.6751, Loss: 0.6340375542640686\n",
      "Epoch: 110, Train: 0.7273, Val: 0.6872, Test: 0.6859, Loss: 0.6319077014923096\n",
      "Epoch: 111, Train: 0.7214, Val: 0.6930, Test: 0.6834, Loss: 0.6291777491569519\n",
      "Epoch: 112, Train: 0.7236, Val: 0.6964, Test: 0.6804, Loss: 0.6291670799255371\n",
      "Epoch: 113, Train: 0.7216, Val: 0.6905, Test: 0.6906, Loss: 0.6245824098587036\n",
      "Epoch: 114, Train: 0.7341, Val: 0.7010, Test: 0.6870, Loss: 0.6234310865402222\n",
      "Epoch: 115, Train: 0.7329, Val: 0.6944, Test: 0.6846, Loss: 0.6302322745323181\n",
      "Epoch: 116, Train: 0.7348, Val: 0.6912, Test: 0.6887, Loss: 0.6199183464050293\n",
      "Epoch: 117, Train: 0.7303, Val: 0.7013, Test: 0.6913, Loss: 0.6118435859680176\n",
      "Epoch: 118, Train: 0.7309, Val: 0.6962, Test: 0.6942, Loss: 0.6140828132629395\n",
      "Epoch: 119, Train: 0.7484, Val: 0.7002, Test: 0.6919, Loss: 0.6227415800094604\n",
      "Epoch: 120, Train: 0.7453, Val: 0.7040, Test: 0.6999, Loss: 0.6113901734352112\n",
      "Epoch: 121, Train: 0.7495, Val: 0.7002, Test: 0.6947, Loss: 0.602495551109314\n",
      "Epoch: 122, Train: 0.7600, Val: 0.6991, Test: 0.6983, Loss: 0.618902325630188\n",
      "Epoch: 123, Train: 0.7567, Val: 0.7064, Test: 0.6970, Loss: 0.6114010810852051\n",
      "Epoch: 124, Train: 0.7646, Val: 0.6991, Test: 0.7056, Loss: 0.6030179262161255\n",
      "Epoch: 125, Train: 0.7748, Val: 0.7139, Test: 0.6992, Loss: 0.5940271615982056\n",
      "Epoch: 126, Train: 0.7757, Val: 0.7049, Test: 0.7028, Loss: 0.5972225666046143\n",
      "Epoch: 127, Train: 0.7821, Val: 0.7090, Test: 0.7044, Loss: 0.6002436876296997\n",
      "Epoch: 128, Train: 0.7796, Val: 0.7080, Test: 0.7075, Loss: 0.5894244313240051\n",
      "Epoch: 129, Train: 0.7974, Val: 0.7029, Test: 0.7149, Loss: 0.5867331624031067\n",
      "Epoch: 130, Train: 0.7943, Val: 0.7138, Test: 0.7095, Loss: 0.5878416299819946\n",
      "Epoch: 131, Train: 0.7880, Val: 0.7107, Test: 0.7161, Loss: 0.5861467719078064\n",
      "Epoch: 132, Train: 0.7986, Val: 0.7167, Test: 0.7164, Loss: 0.5873658061027527\n",
      "Epoch: 133, Train: 0.8216, Val: 0.7288, Test: 0.7223, Loss: 0.5812371969223022\n",
      "Epoch: 134, Train: 0.8057, Val: 0.7222, Test: 0.7204, Loss: 0.5782673358917236\n",
      "Epoch: 135, Train: 0.8056, Val: 0.7182, Test: 0.7264, Loss: 0.57448410987854\n",
      "Epoch: 136, Train: 0.8115, Val: 0.7209, Test: 0.7283, Loss: 0.5755914449691772\n",
      "Epoch: 137, Train: 0.8212, Val: 0.7306, Test: 0.7300, Loss: 0.568964421749115\n",
      "Epoch: 138, Train: 0.8131, Val: 0.7324, Test: 0.7351, Loss: 0.5630272030830383\n",
      "Epoch: 139, Train: 0.8253, Val: 0.7302, Test: 0.7393, Loss: 0.5733718276023865\n",
      "Epoch: 140, Train: 0.8344, Val: 0.7357, Test: 0.7348, Loss: 0.5480000972747803\n",
      "Epoch: 141, Train: 0.8226, Val: 0.7378, Test: 0.7357, Loss: 0.5719251036643982\n",
      "Epoch: 142, Train: 0.8272, Val: 0.7351, Test: 0.7389, Loss: 0.5578988194465637\n",
      "Epoch: 143, Train: 0.8277, Val: 0.7426, Test: 0.7423, Loss: 0.5580206513404846\n",
      "Epoch: 144, Train: 0.8355, Val: 0.7346, Test: 0.7416, Loss: 0.5511246919631958\n",
      "Epoch: 145, Train: 0.8315, Val: 0.7424, Test: 0.7452, Loss: 0.5601673722267151\n",
      "Epoch: 146, Train: 0.8374, Val: 0.7434, Test: 0.7518, Loss: 0.5524490475654602\n",
      "Epoch: 147, Train: 0.8386, Val: 0.7468, Test: 0.7530, Loss: 0.5496705174446106\n",
      "Epoch: 148, Train: 0.8398, Val: 0.7483, Test: 0.7548, Loss: 0.5535192489624023\n",
      "Epoch: 149, Train: 0.8380, Val: 0.7497, Test: 0.7502, Loss: 0.5429545640945435\n",
      "Epoch: 150, Train: 0.8429, Val: 0.7547, Test: 0.7549, Loss: 0.5414234399795532\n",
      "Epoch: 151, Train: 0.8427, Val: 0.7499, Test: 0.7592, Loss: 0.5473601222038269\n",
      "Epoch: 152, Train: 0.8531, Val: 0.7538, Test: 0.7602, Loss: 0.5434921979904175\n",
      "Epoch: 153, Train: 0.8527, Val: 0.7492, Test: 0.7534, Loss: 0.5484768748283386\n",
      "Epoch: 154, Train: 0.8515, Val: 0.7603, Test: 0.7606, Loss: 0.5472995638847351\n",
      "Epoch: 155, Train: 0.8573, Val: 0.7501, Test: 0.7580, Loss: 0.5451675653457642\n",
      "Epoch: 156, Train: 0.8593, Val: 0.7525, Test: 0.7597, Loss: 0.5359841585159302\n",
      "Epoch: 157, Train: 0.8581, Val: 0.7567, Test: 0.7553, Loss: 0.5373581647872925\n",
      "Epoch: 158, Train: 0.8486, Val: 0.7511, Test: 0.7621, Loss: 0.5387391448020935\n",
      "Epoch: 159, Train: 0.8654, Val: 0.7611, Test: 0.7649, Loss: 0.5470179915428162\n",
      "Epoch: 160, Train: 0.8594, Val: 0.7546, Test: 0.7667, Loss: 0.5335506200790405\n",
      "Epoch: 161, Train: 0.8629, Val: 0.7581, Test: 0.7645, Loss: 0.5329647064208984\n",
      "Epoch: 162, Train: 0.8660, Val: 0.7591, Test: 0.7638, Loss: 0.5333133339881897\n",
      "Epoch: 163, Train: 0.8748, Val: 0.7594, Test: 0.7630, Loss: 0.5374255776405334\n",
      "Epoch: 164, Train: 0.8624, Val: 0.7496, Test: 0.7642, Loss: 0.529807984828949\n",
      "Epoch: 165, Train: 0.8734, Val: 0.7489, Test: 0.7634, Loss: 0.5224674940109253\n",
      "Epoch: 166, Train: 0.8728, Val: 0.7480, Test: 0.7645, Loss: 0.5336230993270874\n",
      "Epoch: 167, Train: 0.8786, Val: 0.7455, Test: 0.7619, Loss: 0.525019109249115\n",
      "Epoch: 168, Train: 0.8853, Val: 0.7500, Test: 0.7598, Loss: 0.5334876775741577\n",
      "Epoch: 169, Train: 0.8765, Val: 0.7533, Test: 0.7624, Loss: 0.5238358974456787\n",
      "Epoch: 170, Train: 0.8860, Val: 0.7481, Test: 0.7704, Loss: 0.5290911793708801\n",
      "Epoch: 171, Train: 0.8760, Val: 0.7526, Test: 0.7656, Loss: 0.525763988494873\n",
      "Epoch: 172, Train: 0.8910, Val: 0.7485, Test: 0.7638, Loss: 0.5192937254905701\n",
      "Epoch: 173, Train: 0.8819, Val: 0.7538, Test: 0.7619, Loss: 0.5080286264419556\n",
      "Epoch: 174, Train: 0.8845, Val: 0.7511, Test: 0.7669, Loss: 0.5128852725028992\n",
      "Epoch: 175, Train: 0.8908, Val: 0.7518, Test: 0.7635, Loss: 0.5214087963104248\n",
      "Epoch: 176, Train: 0.8892, Val: 0.7526, Test: 0.7664, Loss: 0.524811327457428\n",
      "Epoch: 177, Train: 0.8817, Val: 0.7475, Test: 0.7640, Loss: 0.5133451223373413\n",
      "Epoch: 178, Train: 0.8893, Val: 0.7529, Test: 0.7756, Loss: 0.5244877934455872\n",
      "Epoch: 179, Train: 0.9001, Val: 0.7596, Test: 0.7675, Loss: 0.5103925466537476\n",
      "Epoch: 180, Train: 0.8893, Val: 0.7610, Test: 0.7726, Loss: 0.5129244923591614\n",
      "Epoch: 181, Train: 0.8963, Val: 0.7550, Test: 0.7691, Loss: 0.5072811841964722\n",
      "Epoch: 182, Train: 0.8960, Val: 0.7617, Test: 0.7754, Loss: 0.5204957127571106\n",
      "Epoch: 183, Train: 0.8948, Val: 0.7513, Test: 0.7769, Loss: 0.5169883370399475\n",
      "Epoch: 184, Train: 0.8974, Val: 0.7510, Test: 0.7717, Loss: 0.5126349925994873\n",
      "Epoch: 185, Train: 0.9049, Val: 0.7533, Test: 0.7769, Loss: 0.5182223320007324\n",
      "Epoch: 186, Train: 0.9011, Val: 0.7478, Test: 0.7747, Loss: 0.5112978219985962\n",
      "Epoch: 187, Train: 0.9003, Val: 0.7607, Test: 0.7704, Loss: 0.5082385540008545\n",
      "Epoch: 188, Train: 0.9062, Val: 0.7604, Test: 0.7777, Loss: 0.5036260485649109\n",
      "Epoch: 189, Train: 0.9056, Val: 0.7513, Test: 0.7745, Loss: 0.5106500387191772\n",
      "Epoch: 190, Train: 0.9033, Val: 0.7536, Test: 0.7699, Loss: 0.5126596093177795\n",
      "Epoch: 191, Train: 0.9087, Val: 0.7549, Test: 0.7716, Loss: 0.5017136931419373\n",
      "Epoch: 192, Train: 0.9071, Val: 0.7601, Test: 0.7758, Loss: 0.4995381534099579\n",
      "Epoch: 193, Train: 0.9082, Val: 0.7540, Test: 0.7776, Loss: 0.49793651700019836\n",
      "Epoch: 194, Train: 0.9052, Val: 0.7629, Test: 0.7734, Loss: 0.502060055732727\n",
      "Epoch: 195, Train: 0.9150, Val: 0.7510, Test: 0.7781, Loss: 0.5101616382598877\n",
      "Epoch: 196, Train: 0.9151, Val: 0.7635, Test: 0.7696, Loss: 0.5020030736923218\n",
      "Epoch: 197, Train: 0.9106, Val: 0.7521, Test: 0.7766, Loss: 0.5057957768440247\n",
      "Epoch: 198, Train: 0.9048, Val: 0.7584, Test: 0.7768, Loss: 0.5016452670097351\n",
      "Epoch: 199, Train: 0.9161, Val: 0.7558, Test: 0.7777, Loss: 0.49302494525909424\n",
      "Train: 0.9135, Val: 0.7562, Test: 0.7745\n"
     ]
    }
   ],
   "source": [
    "pyg_dataset = Planetoid('./tmp/cora', 'Cora')\n",
    "graphs = GraphDataset.pyg_to_graphs(pyg_dataset)\n",
    "\n",
    "dataset = GraphDataset(\n",
    "        graphs,\n",
    "        task='link_pred',\n",
    "        edge_train_mode=\"disjoint\"\n",
    "    )\n",
    "datasets = {}\n",
    "datasets['train'], datasets['val'], datasets['test']= dataset.split(\n",
    "            transductive=True, split_ratio=[0.85, 0.05, 0.1])\n",
    "input_dim = datasets['train'].num_node_features\n",
    "num_classes = datasets['train'].num_edge_labels\n",
    "\n",
    "model = LinkPredModel(input_dim, args[\"hidden_dim\"], num_classes).to(args[\"device\"])\n",
    "model.reset_parameters()\n",
    "\n",
    "optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.9, weight_decay=5e-4)\n",
    "\n",
    "dataloaders = {split: DataLoader(\n",
    "            ds, collate_fn=Batch.collate([]),\n",
    "            batch_size=1, shuffle=(split=='train'))\n",
    "            for split, ds in datasets.items()}\n",
    "best_model = train(model, dataloaders, optimizer, args)\n",
    "log = \"Train: {:.4f}, Val: {:.4f}, Test: {:.4f}\"\n",
    "best_train_roc = test(best_model, dataloaders['train'], args)\n",
    "best_val_roc = test(best_model, dataloaders['val'], args)\n",
    "best_test_roc = test(best_model, dataloaders['test'], args)\n",
    "print(log.format(best_train_roc, best_val_roc, best_test_roc))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "j5brlsKElP0_"
   },
   "source": [
    "## Question 4: What is the maximum ROC-AUC score you could get for the best_model on test set? (13 points)\n",
    "\n",
    "Submit your answers on Gradescope."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "e7JXsMTBgeOI"
   },
   "source": [
    "# Submission\n",
    "\n",
    "In order to get credit, you must go submit your answers on Gradescope.\n",
    "\n",
    "Also, you need to submit the `ipynb` file of Colab 3, by clicking `File` and `Download .ipynb`. Please make sure that your output of each cell is available in your `ipynb` file."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "collapsed_sections": [],
   "machine_shape": "hm",
   "name": "CS224W - Colab 3.ipynb",
   "provenance": [],
   "toc_visible": true
  },
  "kernelspec": {
   "display_name": "Python [conda env:Graph_env] *",
   "language": "python",
   "name": "conda-env-Graph_env-py"
  },
  "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": 1
}