2 changed files with 250 additions and 284 deletions
+ 242
- 283
.ipynb_checkpoints/tensorboarded-checkpoint.ipynb
View File
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "def Graph_load_batch(min_num_nodes=20, max_num_nodes=1000, name='ENZYMES', node_attributes=True, graph_labels=True):\n", | "def Graph_load_batch(min_num_nodes=20, max_num_nodes=1000, name='ENZYMES', node_attributes=True, graph_labels=True):\n", | ||||
| " '''\n", | |||||
| " load many graphs, e.g. enzymes\n", | |||||
| " :return: a list of graphs\n", | |||||
| " '''\n", | |||||
| " print('Loading graph dataset: ' + str(name))\n", | " print('Loading graph dataset: ' + str(name))\n", | ||||
| " G = nx.Graph()\n", | " G = nx.Graph()\n", | ||||
| " # load data\n", | " # load data\n", | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | |||||
| "execution_count": null, | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | "metadata": {}, | ||||
| "outputs": [], | |||||
| "source": [ | "source": [ | ||||
| "Constructing" | |||||
| "Constructing feature matrix of a graph" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "def feature_matrix(g, max_nodes=40):\n", | "def feature_matrix(g, max_nodes=40):\n", | ||||
| " esm = nx.get_node_attributes(g, 'label')\n", | |||||
| " piazche = np.zeros((max_nodes, 3))\n", | |||||
| " for i, (k, v) in enumerate(esm.items()):\n", | |||||
| " # print(i, k , v)\n", | |||||
| " piazche[i][v-1] = 1\n", | |||||
| " return piazche" | |||||
| " esm = nx.get_node_attributes(g, 'label')\n", | |||||
| " piazche = np.zeros((max_nodes, 3))\n", | |||||
| " for i, (k, v) in enumerate(esm.items()):\n", | |||||
| " # print(i, k , v)\n", | |||||
| " piazche[i][v-1] = 1\n", | |||||
| " return piazche" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Removing a random node from a graph\n", | |||||
| "\n", | |||||
| "Returns remaining graph with removed node links" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "code", | |||||
| "execution_count": 12, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "PkeAYLDQ2W5D" | |||||
| }, | |||||
| "outputs": [], | |||||
| "source": [ | |||||
| "def remove_random_node(graph, max_size=40, min_size=10):\n", | |||||
| " if len(graph.nodes()) >= max_size or len(graph.nodes()) < min_size:\n", | |||||
| " return None\n", | |||||
| " relabeled_graph = nx.relabel.convert_node_labels_to_integers(graph)\n", | |||||
| " choice = np.random.choice(list(relabeled_graph.nodes()))\n", | |||||
| " remaining_graph = nx.to_numpy_matrix(relabeled_graph.subgraph(filter(lambda x: x != choice, list(relabeled_graph.nodes()))))\n", | |||||
| " removed_node = nx.to_numpy_matrix(relabeled_graph)[choice]\n", | |||||
| " graph_length = len(remaining_graph)\n", | |||||
| " source_graph = np.pad(remaining_graph, [(0, max_size - graph_length), (0, max_size - graph_length)])\n", | |||||
| " # target_graph = np.copy(source_graph)\n", | |||||
| " removed_node_row = np.asarray(removed_node)[0]\n", | |||||
| " # target_graph[graph_length] = np.pad(removed_node_row, [(0, max_size - len(removed_node_row))])\n", | |||||
| " return remaining_graph, removed_node_row" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Prepare graphs for the model\n", | |||||
| "\n", | |||||
| "returns a graph with a randomly removed node adj matrix [0], its feature matrix [1], the removed node true links [2] " | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "code", | |||||
| "execution_count": 13, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "FXFQJIvE2Ync" | |||||
| }, | |||||
| "outputs": [], | |||||
| "source": [ | |||||
| "def prepare_graph_data(graph, max_size=40, min_size=10):\n", | |||||
| " if len(graph.nodes()) >= max_size or len(graph.nodes()) < min_size:\n", | |||||
| " return None\n", | |||||
| " relabeled_graph = nx.relabel.convert_node_labels_to_integers(graph)\n", | |||||
| " choice = np.random.choice(list(relabeled_graph.nodes()))\n", | |||||
| " remaining_graph = relabeled_graph.subgraph(filter(lambda x: x != choice, list(relabeled_graph.nodes())))\n", | |||||
| " remaining_graph_adj = nx.to_numpy_matrix(remaining_graph)\n", | |||||
| " graph_length = len(remaining_graph)\n", | |||||
| " remaining_graph_adj = np.pad(remaining_graph_adj, [(0, max_size - graph_length), (0, max_size - graph_length)])\n", | |||||
| " removed_node = nx.to_numpy_matrix(relabeled_graph)[choice]\n", | |||||
| " removed_node_row = np.asarray(removed_node)[0]\n", | |||||
| " removed_node_row = np.pad(removed_node_row, [(0, max_size - len(removed_node_row))])\n", | |||||
| " return remaining_graph_adj, feature_matrix(remaining_graph), removed_node_row" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | "cell_type": "code", | ||||
| "execution_count": 10, | |||||
| "execution_count": 30, | |||||
| "metadata": { | "metadata": { | ||||
| "colab": { | |||||
| "base_uri": "https://localhost:8080/", | |||||
| "height": 34 | |||||
| }, | |||||
| "colab": {}, | |||||
| "colab_type": "code", | "colab_type": "code", | ||||
| "id": "KdO6zZruosHe", | |||||
| "outputId": "e226f7eb-b285-4e02-b576-2b2998c9240f" | |||||
| "id": "tQdkjvY22_Kf" | |||||
| }, | }, | ||||
| "outputs": [ | |||||
| { | |||||
| "data": { | |||||
| "text/plain": [ | |||||
| "469" | |||||
| ] | |||||
| }, | |||||
| "execution_count": 10, | |||||
| "metadata": { | |||||
| "tags": [] | |||||
| }, | |||||
| "output_type": "execute_result" | |||||
| } | |||||
| ], | |||||
| "outputs": [], | |||||
| "source": [ | "source": [ | ||||
| "# len(train)" | |||||
| "# coop = sum([list(filter(lambda x: x is not None, [prepare_graph_data(g) for g in graphs])) for i in range(10)], [])\n", | |||||
| "coop = list(filter(lambda x: x is not None, [prepare_graph_data(g) for g in train]))\n", | |||||
| "dale = list(filter(lambda x: x is not None, [prepare_graph_data(g) for g in test]))" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | "cell_type": "code", | ||||
| "execution_count": 12, | |||||
| "execution_count": 47, | |||||
| "metadata": { | "metadata": { | ||||
| "colab": {}, | "colab": {}, | ||||
| "colab_type": "code", | "colab_type": "code", | ||||
| "id": "PkeAYLDQ2W5D" | |||||
| "id": "qChhpZuCpHWv" | |||||
| }, | }, | ||||
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "def remove_random_node(graph, max_size=40, min_size=10):\n", | |||||
| " if len(graph.nodes()) >= max_size or len(graph.nodes()) < min_size:\n", | |||||
| " return None\n", | |||||
| " relabeled_graph = nx.relabel.convert_node_labels_to_integers(graph)\n", | |||||
| " choice = np.random.choice(list(relabeled_graph.nodes()))\n", | |||||
| " remaining_graph = nx.to_numpy_matrix(relabeled_graph.subgraph(filter(lambda x: x != choice, list(relabeled_graph.nodes()))))\n", | |||||
| " removed_node = nx.to_numpy_matrix(relabeled_graph)[choice]\n", | |||||
| " graph_length = len(remaining_graph)\n", | |||||
| " source_graph = np.pad(remaining_graph, [(0, max_size - graph_length), (0, max_size - graph_length)])\n", | |||||
| " # target_graph = np.copy(source_graph)\n", | |||||
| " removed_node_row = np.asarray(removed_node)[0]\n", | |||||
| " # target_graph[graph_length] = np.pad(removed_node_row, [(0, max_size - len(removed_node_row))])\n", | |||||
| " return remaining_graph, removed_node_row" | |||||
| "trainloader = torch.utils.data.DataLoader(coop, collate_fn=lambda x: x[0], batch_size=1)" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | "cell_type": "code", | ||||
| "execution_count": 13, | |||||
| "execution_count": 32, | |||||
| "metadata": { | "metadata": { | ||||
| "colab": {}, | "colab": {}, | ||||
| "colab_type": "code", | "colab_type": "code", | ||||
| "id": "FXFQJIvE2Ync" | |||||
| "id": "8FHvrVFlqNh2" | |||||
| }, | }, | ||||
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "def prepare_graph_data(graph, max_size=40, min_size=10):\n", | |||||
| " '''\n", | |||||
| " gets a graph as an input\n", | |||||
| " returns a graph with a randomly removed node adj matrix [0], its feature matrix [1], the removed node true links [2] \n", | |||||
| " '''\n", | |||||
| " if len(graph.nodes()) >= max_size or len(graph.nodes()) < min_size:\n", | |||||
| " return None\n", | |||||
| " relabeled_graph = nx.relabel.convert_node_labels_to_integers(graph)\n", | |||||
| " choice = np.random.choice(list(relabeled_graph.nodes()))\n", | |||||
| " remaining_graph = relabeled_graph.subgraph(filter(lambda x: x != choice, list(relabeled_graph.nodes())))\n", | |||||
| " remaining_graph_adj = nx.to_numpy_matrix(remaining_graph)\n", | |||||
| " graph_length = len(remaining_graph)\n", | |||||
| " remaining_graph_adj = np.pad(remaining_graph_adj, [(0, max_size - graph_length), (0, max_size - graph_length)])\n", | |||||
| " removed_node = nx.to_numpy_matrix(relabeled_graph)[choice]\n", | |||||
| " removed_node_row = np.asarray(removed_node)[0]\n", | |||||
| " removed_node_row = np.pad(removed_node_row, [(0, max_size - len(removed_node_row))])\n", | |||||
| " return remaining_graph_adj, feature_matrix(remaining_graph), removed_node_row" | |||||
| "testloader = torch.utils.data.DataLoader(dale, collate_fn=lambda x: x[0], batch_size=1)" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "id": "JEvec2nosVn7" | "id": "JEvec2nosVn7" | ||||
| }, | }, | ||||
| "source": [ | "source": [ | ||||
| "# Model" | |||||
| "# Building Model" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Graph convolutional layer for extracting initial features" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| " return y" | " return y" | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Attention calculator using given key, query and value" | |||||
| ] | |||||
| }, | |||||
| { | { | ||||
| "cell_type": "code", | "cell_type": "code", | ||||
| "execution_count": 15, | "execution_count": 15, | ||||
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "def attention(query, key, value, key_dim):\n", | "def attention(query, key, value, key_dim):\n", | ||||
| " # print('key:', key.transpose(-2,-1))\n", | |||||
| " scores = torch.matmul(query, key.transpose(-2,-1)) / math.sqrt(key_dim)\n", | |||||
| " scores = torch.matmul(scores, value)\n", | |||||
| " # print('scores:', scores)\n", | |||||
| " scores = F.softmax(scores)\n", | |||||
| " # scores = torch.sigmoid(scores)\n", | |||||
| " return scores" | |||||
| " # print('key:', key.transpose(-2,-1))\n", | |||||
| " scores = torch.matmul(query, key.transpose(-2,-1)) / math.sqrt(key_dim)\n", | |||||
| " scores = torch.matmul(scores, value)\n", | |||||
| " # print('scores:', scores)\n", | |||||
| " scores = F.softmax(scores)\n", | |||||
| " # scores = torch.sigmoid(scores)\n", | |||||
| " return scores" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Graph attention layer for more features" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "class GraphAttn(nn.Module):\n", | "class GraphAttn(nn.Module):\n", | ||||
| " def __init__(self, heads, model_dim, dropout=0.1):\n", | |||||
| " super().__init__()\n", | |||||
| " self.model_dim = model_dim\n", | |||||
| " self.key_dim = model_dim // heads\n", | |||||
| " self.heads = heads\n", | |||||
| " def __init__(self, heads, model_dim, dropout=0.1):\n", | |||||
| " super().__init__()\n", | |||||
| " self.model_dim = model_dim\n", | |||||
| " self.key_dim = model_dim // heads\n", | |||||
| " self.heads = heads\n", | |||||
| "\n", | "\n", | ||||
| " self.q_linear = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| " self.v_linear = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| " self.k_linear = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| " self.q_linear = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| " self.v_linear = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| " self.k_linear = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| "\n", | "\n", | ||||
| " self.dropout = nn.Dropout(dropout)\n", | |||||
| " self.out = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| " self.dropout = nn.Dropout(dropout)\n", | |||||
| " self.out = nn.Linear(model_dim, model_dim).cuda()\n", | |||||
| "\n", | "\n", | ||||
| " def forward(self, query, key, value):\n", | |||||
| " # print(q, k, v)\n", | |||||
| " bs = query.size(0)\n", | |||||
| " def forward(self, query, key, value):\n", | |||||
| " # print(q, k, v)\n", | |||||
| " bs = query.size(0)\n", | |||||
| "\n", | "\n", | ||||
| " key = self.k_linear(key.float()).view(bs, -1, self.heads, self.key_dim)\n", | |||||
| " query = self.q_linear(query.float()).view(bs, -1, self.heads, self.key_dim)\n", | |||||
| " value = self.v_linear(value.float()).view(bs, -1, self.heads, self.key_dim)\n", | |||||
| " \n", | |||||
| " key = key.transpose(1,2)\n", | |||||
| " query = query.transpose(1,2)\n", | |||||
| " value = value.transpose(1,2)\n", | |||||
| " key = self.k_linear(key.float()).view(bs, -1, self.heads, self.key_dim)\n", | |||||
| " query = self.q_linear(query.float()).view(bs, -1, self.heads, self.key_dim)\n", | |||||
| " value = self.v_linear(value.float()).view(bs, -1, self.heads, self.key_dim)\n", | |||||
| "\n", | "\n", | ||||
| " scores = attention(query, key, value, self.key_dim)\n", | |||||
| " concat = scores.transpose(1,2).contiguous().view(bs, -1, self.model_dim)\n", | |||||
| " output = self.out(concat)\n", | |||||
| " output = output.view(bs, self.model_dim)\n", | |||||
| " key = key.transpose(1,2)\n", | |||||
| " query = query.transpose(1,2)\n", | |||||
| " value = value.transpose(1,2)\n", | |||||
| "\n", | "\n", | ||||
| " return output" | |||||
| " scores = attention(query, key, value, self.key_dim)\n", | |||||
| " concat = scores.transpose(1,2).contiguous().view(bs, -1, self.model_dim)\n", | |||||
| " output = self.out(concat)\n", | |||||
| " output = output.view(bs, self.model_dim)\n", | |||||
| "\n", | |||||
| " return output" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "MLP layer to map features to links" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| " return output" | " return output" | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Assembeled model" | |||||
| ] | |||||
| }, | |||||
| { | { | ||||
| "cell_type": "code", | "cell_type": "code", | ||||
| "execution_count": 18, | "execution_count": 18, | ||||
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "class Hydra(nn.Module):\n", | "class Hydra(nn.Module):\n", | ||||
| " def __init__(self, gcn_input, model_dim, head):\n", | |||||
| " super().__init__()\n", | |||||
| " def __init__(self, gcn_input, model_dim, head):\n", | |||||
| " super().__init__()\n", | |||||
| "\n", | "\n", | ||||
| " self.GCN = GraphConv(input_dim=gcn_input, output_dim=model_dim).cuda()\n", | |||||
| " self.GAT = GraphAttn(heads=head, model_dim=model_dim).cuda()\n", | |||||
| " self.MLP = FeedForward(input_size=model_dim, hidden_size=gcn_input).cuda()\n", | |||||
| " self.GCN = GraphConv(input_dim=gcn_input, output_dim=model_dim).cuda()\n", | |||||
| " self.GAT = GraphAttn(heads=head, model_dim=model_dim).cuda()\n", | |||||
| " self.MLP = FeedForward(input_size=model_dim, hidden_size=gcn_input).cuda()\n", | |||||
| "\n", | "\n", | ||||
| " def forward(self, x, adj):\n", | |||||
| " gcn_outputs = self.GCN(x, adj)\n", | |||||
| " gat_output = self.GAT(gcn_outputs, gcn_outputs, gcn_outputs)\n", | |||||
| " mlp_output = self.MLP(gat_output).reshape(1,-1)\n", | |||||
| " def forward(self, x, adj):\n", | |||||
| " gcn_outputs = self.GCN(x, adj)\n", | |||||
| " gat_output = self.GAT(gcn_outputs, gcn_outputs, gcn_outputs)\n", | |||||
| " mlp_output = self.MLP(gat_output).reshape(1,-1)\n", | |||||
| "\n", | "\n", | ||||
| " return mlp_output" | |||||
| " return mlp_output" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Building model with given inputs" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | |||||
| "execution_count": null, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "hKeg1-8P2kqK" | |||||
| }, | |||||
| "outputs": [], | |||||
| "source": [ | |||||
| "# adj, features, true_links = prepare_graph_data(graphs[0])" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "code", | |||||
| "execution_count": null, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "8QgGRjyt2sFN" | |||||
| }, | |||||
| "outputs": [], | |||||
| "source": [ | |||||
| "# adj, features, true_links = torch.tensor(adj).cuda(), torch.tensor(features).cuda(), torch.tensor(true_links).cuda()" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "code", | |||||
| "execution_count": null, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "tcSaiDch3CUG" | |||||
| }, | |||||
| "outputs": [], | |||||
| "source": [ | |||||
| "# kyle = build_model(3, 243, 9)" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "code", | |||||
| "execution_count": null, | |||||
| "metadata": { | |||||
| "colab": { | |||||
| "base_uri": "https://localhost:8080/", | |||||
| "height": 173 | |||||
| }, | |||||
| "colab_type": "code", | |||||
| "id": "tjioYsvs2tUR", | |||||
| "outputId": "06033c55-7e06-4d6c-8b98-a4bbff7bf9f2" | |||||
| }, | |||||
| "outputs": [ | |||||
| { | |||||
| "name": "stderr", | |||||
| "output_type": "stream", | |||||
| "text": [ | |||||
| "/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:6: UserWarning: Implicit dimension choice for softmax has been deprecated. Change the call to include dim=X as an argument.\n", | |||||
| " \n" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "data": { | |||||
| "text/plain": [ | |||||
| "tensor([[0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425,\n", | |||||
| " 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425,\n", | |||||
| " 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425, 0.5425,\n", | |||||
| " 0.5425, 0.5425, 0.5425, 0.5548, 0.5548, 0.5425, 0.5548, 0.5425, 0.5548,\n", | |||||
| " 0.5548, 0.5548, 0.5548, 0.5548]], device='cuda:0',\n", | |||||
| " grad_fn=<ViewBackward>)" | |||||
| ] | |||||
| }, | |||||
| "execution_count": 20, | |||||
| "metadata": { | |||||
| "tags": [] | |||||
| }, | |||||
| "output_type": "execute_result" | |||||
| } | |||||
| ], | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | "source": [ | ||||
| "# kyle(features, adj)" | |||||
| "# Evaluating Functions" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "id": "FxNdr2zR3Zgo" | "id": "FxNdr2zR3Zgo" | ||||
| }, | }, | ||||
| "source": [ | "source": [ | ||||
| "# Hala Train" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "code", | |||||
| "execution_count": 23, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "t_k4B_3ko8ps" | |||||
| }, | |||||
| "outputs": [], | |||||
| "source": [ | |||||
| "def fn(batch):\n", | |||||
| " return batch[0]" | |||||
| "# Training Model" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | |||||
| "execution_count": 30, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "tQdkjvY22_Kf" | |||||
| }, | |||||
| "outputs": [], | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | "source": [ | ||||
| "# coop = sum([list(filter(lambda x: x is not None, [prepare_graph_data(g) for g in graphs])) for i in range(10)], [])\n", | |||||
| "coop = list(filter(lambda x: x is not None, [prepare_graph_data(g) for g in train]))\n", | |||||
| "dale = list(filter(lambda x: x is not None, [prepare_graph_data(g) for g in test]))" | |||||
| "Training the model with given data and number of epochs" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | "cell_type": "code", | ||||
| "execution_count": 47, | |||||
| "execution_count": 57, | |||||
| "metadata": { | "metadata": { | ||||
| "colab": {}, | "colab": {}, | ||||
| "colab_type": "code", | "colab_type": "code", | ||||
| "id": "qChhpZuCpHWv" | |||||
| "id": "PEXlurWtpII2" | |||||
| }, | }, | ||||
| "outputs": [], | "outputs": [], | ||||
| "source": [ | "source": [ | ||||
| "trainloader_train = torch.utils.data.DataLoader(coop, collate_fn=fn, batch_size=1)" | |||||
| "def train_model(model, trainloader, epoch, print_every=100):\n", | |||||
| " optim = torch.optim.Adam(model.parameters(), lr=0.001, betas=(0.9, 0.98), eps=1e-9)\n", | |||||
| "\n", | |||||
| " model.train()\n", | |||||
| " start = time.time()\n", | |||||
| " temp = start\n", | |||||
| " total_loss = 0\n", | |||||
| "\n", | |||||
| " for i in range(epoch):\n", | |||||
| " for batch, data in enumerate(trainloader, 0):\n", | |||||
| " adj, features, true_links = data\n", | |||||
| " adj, features, true_links = torch.tensor(adj).cuda(), torch.tensor(features).cuda(), torch.tensor(true_links).cuda()\n", | |||||
| " # print(adj.shape)\n", | |||||
| " # print(features.shape)\n", | |||||
| " # print(true_links.shape)\n", | |||||
| " preds = model(features, adj)\n", | |||||
| " optim.zero_grad()\n", | |||||
| " loss = F.binary_cross_entropy(preds.double(), true_links.double())\n", | |||||
| " writer.add_scalar('Loss/train', float(loss), i)\n", | |||||
| " loss.backward()\n", | |||||
| " optim.step()\n", | |||||
| " total_loss += loss.item()\n", | |||||
| " if (i + 1) % print_every == 0:\n", | |||||
| " loss_avg = total_loss / print_every\n", | |||||
| " print(\"time = %dm, epoch %d, iter = %d, loss = %.3f,\\\n", | |||||
| " %ds per %d iters\" % ((time.time() - start) // 60,\\\n", | |||||
| " epoch + 1, i + 1, loss_avg, time.time() - temp,\\\n", | |||||
| " print_every))\n", | |||||
| " total_loss = 0\n", | |||||
| " temp = time.time()" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "kyle = build_model(3, 243, 9)" | "kyle = build_model(3, 243, 9)" | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | |||||
| "cell_type": "code", | |||||
| "execution_count": 57, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "PEXlurWtpII2" | |||||
| }, | |||||
| "outputs": [], | |||||
| "source": [ | |||||
| "def train_model(model, trainloader, epoch, print_every=100):\n", | |||||
| " optim = torch.optim.Adam(model.parameters(), lr=0.001, betas=(0.9, 0.98), eps=1e-9)\n", | |||||
| "\n", | |||||
| " model.train()\n", | |||||
| " start = time.time()\n", | |||||
| " temp = start\n", | |||||
| " total_loss = 0\n", | |||||
| "\n", | |||||
| " for i in range(epoch):\n", | |||||
| " for batch, data in enumerate(trainloader, 0):\n", | |||||
| " adj, features, true_links = data\n", | |||||
| " adj, features, true_links = torch.tensor(adj).cuda(), torch.tensor(features).cuda(), torch.tensor(true_links).cuda()\n", | |||||
| " # print(adj.shape)\n", | |||||
| " # print(features.shape)\n", | |||||
| " # print(true_links.shape)\n", | |||||
| " preds = model(features, adj)\n", | |||||
| " optim.zero_grad()\n", | |||||
| " loss = F.binary_cross_entropy(preds.double(), true_links.double())\n", | |||||
| " writer.add_scalar('Loss/train', float(loss), i)\n", | |||||
| " loss.backward()\n", | |||||
| " optim.step()\n", | |||||
| " total_loss += loss.item()\n", | |||||
| " if (i + 1) % print_every == 0:\n", | |||||
| " loss_avg = total_loss / print_every\n", | |||||
| " print(\"time = %dm, epoch %d, iter = %d, loss = %.3f,\\\n", | |||||
| " %ds per %d iters\" % ((time.time() - start) // 60,\\\n", | |||||
| " epoch + 1, i + 1, loss_avg, time.time() - temp,\\\n", | |||||
| " print_every))\n", | |||||
| " total_loss = 0\n", | |||||
| " temp = time.time()" | |||||
| ] | |||||
| }, | |||||
| { | { | ||||
| "cell_type": "code", | "cell_type": "code", | ||||
| "execution_count": 58, | "execution_count": 58, | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "cell_type": "code", | |||||
| "execution_count": 32, | |||||
| "metadata": { | |||||
| "colab": {}, | |||||
| "colab_type": "code", | |||||
| "id": "8FHvrVFlqNh2" | |||||
| }, | |||||
| "outputs": [], | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | "source": [ | ||||
| "trainloader_test = torch.utils.data.DataLoader(dale, collate_fn=fn, batch_size=1)" | |||||
| "# Testing Model" | |||||
| ] | |||||
| }, | |||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "Testing model and printing the loss" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| " # total_loss = 0\n", | " # total_loss = 0\n", | ||||
| "\n", | "\n", | ||||
| " for batch, data in enumerate(trainloader, 0):\n", | " for batch, data in enumerate(trainloader, 0):\n", | ||||
| " adj, features, true_links = data\n", | |||||
| " adj, features, true_links = torch.tensor(adj).cuda(), torch.tensor(features).cuda(), torch.tensor(true_links).cuda()\n", | |||||
| " # print(adj.shape)\n", | |||||
| " # print(features.shape)\n", | |||||
| " # print(true_links.shape)\n", | |||||
| " preds = model(features, adj)\n", | |||||
| " loss = F.binary_cross_entropy(preds.double(), true_links.double())\n", | |||||
| " # total_loss += loss.item()\n", | |||||
| " # loss_avg = total_loss / print_every\n", | |||||
| " if (batch + 1) % print_every == 0:\n", | |||||
| " print(\"loss = \", float(loss))\n", | |||||
| " temp = time.time()" | |||||
| " adj, features, true_links = data\n", | |||||
| " adj, features, true_links = torch.tensor(adj).cuda(), torch.tensor(features).cuda(), torch.tensor(true_links).cuda()\n", | |||||
| " # print(adj.shape)\n", | |||||
| " # print(features.shape)\n", | |||||
| " # print(true_links.shape)\n", | |||||
| " preds = model(features, adj)\n", | |||||
| " loss = F.binary_cross_entropy(preds.double(), true_links.double())\n", | |||||
| " # total_loss += loss.item()\n", | |||||
| " # loss_avg = total_loss / print_every\n", | |||||
| " if (batch + 1) % print_every == 0:\n", | |||||
| " print(\"loss = \", float(loss))\n", | |||||
| " temp = time.time()" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "test_model(kyle, trainloader_test)" | "test_model(kyle, trainloader_test)" | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "# Evaluating Model" | |||||
| ] | |||||
| }, | |||||
| { | { | ||||
| "cell_type": "markdown", | "cell_type": "markdown", | ||||
| "metadata": { | "metadata": { |
+ 8
- 1
tensorboarded.ipynb
View File
| "cell_type": "markdown", | "cell_type": "markdown", | ||||
| "metadata": {}, | "metadata": {}, | ||||
| "source": [ | "source": [ | ||||
| "# Evaluating Model" | |||||
| "# Evaluating Functions" | |||||
| ] | ] | ||||
| }, | }, | ||||
| { | { | ||||
| "test_model(kyle, trainloader_test)" | "test_model(kyle, trainloader_test)" | ||||
| ] | ] | ||||
| }, | }, | ||||
| { | |||||
| "cell_type": "markdown", | |||||
| "metadata": {}, | |||||
| "source": [ | |||||
| "# Evaluating Model" | |||||
| ] | |||||
| }, | |||||
| { | { | ||||
| "cell_type": "markdown", | "cell_type": "markdown", | ||||
| "metadata": { | "metadata": { |
Loading…