basic_LightGCN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os.path as osp\n",
    "import os\n",
    "\n",
    "#torch 전에 할댕해줘야함\n",
    "os.environ[\"CUDA_DEVICE_ORDER\"]=\"PCI_BUS_ID\" \n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"]= \"1\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "import torch\n",
    "from tqdm import tqdm\n",
    "\n",
    "from torch_geometric.datasets import AmazonBook, MovieLens100K\n",
    "from torch_geometric.nn import LightGCN\n",
    "from torch_geometric.utils import degree\n",
    "from torch_geometric.data import InMemoryDataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Device: cuda\n",
      "Current cuda device: 0\n",
      "Count of using GPUs: 1\n"
     ]
    }
   ],
   "source": [
    "\n",
    "\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "print('Device:', device)\n",
    "print('Current cuda device:', torch.cuda.current_device())\n",
    "print('Count of using GPUs:', torch.cuda.device_count())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "tensor([[   5,    9,   11,  ...,  933,    9,  681],\n",
       "        [1682, 1682, 1682,  ..., 2140, 2141, 2143]], device='cuda:0')"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# ml 1m \n",
    "path = osp.join('./', 'data', 'ML1M')\n",
    "dataset = MovieLens100K(path)\n",
    "dataset.process()\n",
    "data = dataset[0]\n",
    "num_users, num_books = data['user'].num_nodes, data['movie'].num_nodes\n",
    "data = data.to_homogeneous().to(device)\n",
    "# 이새끼 호모지니어스로 바꾸면 엣지 라벨 순서가 바뀌어 버림\n",
    "data.edge_label_index = torch.stack([data.edge_label_index[1],data.edge_label_index[0]], dim = 0)\n",
    "data.edge_label_index"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 아마존\n",
    "path = osp.join('./', 'data', 'Amazon')\n",
    "dataset = AmazonBook(path)\n",
    "data = dataset[0]\n",
    "num_users, num_books = data['user'].num_nodes, data['book'].num_nodes\n",
    "data = data.to_homogeneous().to(device)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_3560841/1471277341.py:127: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at /opt/conda/conda-bld/pytorch_1729647327249/work/torch/csrc/utils/tensor_new.cpp:278.)\n",
      "  index = torch.tensor([rows, cols])\n"
     ]
    }
   ],
   "source": [
    "# Gowalla\n",
    "from .datasets import Gowalla\n",
    "path = osp.join('./', 'data', 'Gowalla')\n",
    "dataset = Gowalla(path)\n",
    "dataset.process()\n",
    "data = dataset[0]\n",
    "num_users, num_books = data['user'].num_nodes, data['item'].num_nodes\n",
    "data = data.to_homogeneous().to(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "HeteroData(\n",
       "  user={ num_nodes=29858 },\n",
       "  item={ num_nodes=40988 },\n",
       "  (user, rates, item)={\n",
       "    edge_index=[2, 821971],\n",
       "    edge_label_index=[2, 205493],\n",
       "  },\n",
       "  (item, rated_by, user)={ edge_index=[2, 821971] }\n",
       ")"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Use all message passing edges as training labels:\n",
    "batch_size = 1024\n",
    "mask = data.edge_index[0] < data.edge_index[1]\n",
    "train_edge_label_index = data.edge_index[:, mask]\n",
    "train_loader = torch.utils.data.DataLoader(\n",
    "    range(train_edge_label_index.size(1)),\n",
    "    shuffle=True,\n",
    "    batch_size=batch_size,\n",
    ")\n",
    "\n",
    "model = LightGCN(\n",
    "    num_nodes=data.num_nodes,\n",
    "    embedding_dim=64,\n",
    "    num_layers=3\n",
    ").to(device)\n",
    "optimizer = torch.optim.Adam(model.parameters(), lr=0.001)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train():\n",
    "    total_loss = total_examples = 0\n",
    "\n",
    "    for index in train_loader:\n",
    "        # Sample positive and negative labels.\n",
    "        pos_edge_label_index = train_edge_label_index[:, index]\n",
    "        neg_edge_label_index = torch.stack([\n",
    "            pos_edge_label_index[0],\n",
    "            torch.randint(num_users, num_users + num_books,\n",
    "                          (index.numel(), ), device=device)\n",
    "        ], dim=0)\n",
    "        edge_label_index = torch.cat([\n",
    "            pos_edge_label_index,\n",
    "            neg_edge_label_index,\n",
    "        ], dim=1)\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "        pos_rank, neg_rank = model(data.edge_index, edge_label_index).chunk(2)\n",
    "\n",
    "        loss = model.recommendation_loss(\n",
    "            pos_rank,\n",
    "            neg_rank,\n",
    "            node_id=edge_label_index.unique(),\n",
    "            lambda_reg = 0.0001\n",
    "        )\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "        total_loss += float(loss) * pos_rank.numel()\n",
    "        total_examples += pos_rank.numel()\n",
    "\n",
    "    return total_loss / total_examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "@torch.no_grad()\n",
    "def test(k: int):\n",
    "    emb = model.get_embedding(data.edge_index)\n",
    "    user_emb, book_emb = emb[:num_users], emb[num_users:]\n",
    "\n",
    "    precision = recall = total_examples = 0\n",
    "    for start in range(0, num_users, batch_size):\n",
    "        end = min(start + batch_size, num_users)\n",
    "        logits = user_emb[start:end] @ book_emb.t()\n",
    "        # Exclude training edges:\n",
    "        mask = ((train_edge_label_index[0] >= start) &\n",
    "                (train_edge_label_index[0] < end))\n",
    "        logits[train_edge_label_index[0, mask] - start,\n",
    "                train_edge_label_index[1, mask] - num_users] = float('-inf')\n",
    "\n",
    "        # Computing precision and recall:\n",
    "        ground_truth = torch.zeros_like(logits, dtype=torch.bool)\n",
    "        mask = ((data.edge_label_index[0] >= start) &\n",
    "                (data.edge_label_index[0] < end))\n",
    "        ground_truth[data.edge_label_index[0, mask] - start,\n",
    "                        data.edge_label_index[1, mask] - num_users] = True\n",
    "        node_count = degree(data.edge_label_index[0, mask] - start,\n",
    "                                num_nodes=logits.size(0))\n",
    "\n",
    "        topk_index = logits.topk(k, dim=-1).indices\n",
    "        isin_mat = ground_truth.gather(1, topk_index)\n",
    "\n",
    "        precision += float((isin_mat.sum(dim=-1) / k).sum())\n",
    "        recall += float((isin_mat.sum(dim=-1) / node_count.clamp(1e-6)).sum())\n",
    "        total_examples += int((node_count > 0).sum())\n",
    "\n",
    "    return precision / total_examples, recall / total_examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 1/1000 [00:18<5:05:58, 18.38s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 000, Loss: 0.3528, Precision@20: 0.0209, Recall@20: 0.0702\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  1%|          | 11/1000 [03:23<5:06:18, 18.58s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 010, Loss: 0.0931, Precision@20: 0.0321, Recall@20: 0.1109\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  2%|▏         | 21/1000 [06:31<5:06:26, 18.78s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 020, Loss: 0.0672, Precision@20: 0.0365, Recall@20: 0.1263\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  3%|▎         | 31/1000 [09:39<5:04:15, 18.84s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 030, Loss: 0.0546, Precision@20: 0.0398, Recall@20: 0.1375\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  4%|▍         | 41/1000 [12:47<4:59:13, 18.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 040, Loss: 0.0475, Precision@20: 0.0423, Recall@20: 0.1463\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  5%|▌         | 51/1000 [15:54<4:54:53, 18.64s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 050, Loss: 0.0427, Precision@20: 0.0441, Recall@20: 0.1521\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  6%|▌         | 61/1000 [19:00<4:51:38, 18.63s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 060, Loss: 0.0390, Precision@20: 0.0455, Recall@20: 0.1569\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  7%|▋         | 71/1000 [22:06<4:48:38, 18.64s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 070, Loss: 0.0362, Precision@20: 0.0467, Recall@20: 0.1608\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  8%|▊         | 81/1000 [25:13<4:44:49, 18.60s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 080, Loss: 0.0339, Precision@20: 0.0478, Recall@20: 0.1645\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  9%|▉         | 91/1000 [28:19<4:42:22, 18.64s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 090, Loss: 0.0324, Precision@20: 0.0486, Recall@20: 0.1676\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 10%|█         | 101/1000 [31:25<4:39:36, 18.66s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 100, Loss: 0.0308, Precision@20: 0.0494, Recall@20: 0.1704\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 11%|█         | 111/1000 [34:31<4:35:31, 18.60s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 110, Loss: 0.0299, Precision@20: 0.0500, Recall@20: 0.1723\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 12%|█▏        | 121/1000 [37:37<4:32:31, 18.60s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 120, Loss: 0.0289, Precision@20: 0.0506, Recall@20: 0.1738\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 13%|█▎        | 131/1000 [40:43<4:29:33, 18.61s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 130, Loss: 0.0280, Precision@20: 0.0511, Recall@20: 0.1755\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 14%|█▍        | 141/1000 [43:51<4:27:57, 18.72s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 140, Loss: 0.0275, Precision@20: 0.0516, Recall@20: 0.1774\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 15%|█▌        | 151/1000 [46:57<4:23:54, 18.65s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 150, Loss: 0.0268, Precision@20: 0.0518, Recall@20: 0.1781\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 16%|█▌        | 161/1000 [50:03<4:20:27, 18.63s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 160, Loss: 0.0262, Precision@20: 0.0522, Recall@20: 0.1792\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 17%|█▋        | 171/1000 [53:10<4:17:15, 18.62s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 170, Loss: 0.0257, Precision@20: 0.0526, Recall@20: 0.1810\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 18%|█▊        | 181/1000 [56:17<4:16:27, 18.79s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 180, Loss: 0.0255, Precision@20: 0.0530, Recall@20: 0.1821\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 19%|█▉        | 191/1000 [59:24<4:11:09, 18.63s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 190, Loss: 0.0250, Precision@20: 0.0531, Recall@20: 0.1827\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 20%|██        | 201/1000 [1:02:30<4:07:52, 18.61s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 200, Loss: 0.0248, Precision@20: 0.0534, Recall@20: 0.1837\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 21%|██        | 211/1000 [1:05:37<4:04:58, 18.63s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 210, Loss: 0.0245, Precision@20: 0.0537, Recall@20: 0.1847\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 22%|██▏       | 221/1000 [1:08:43<4:01:42, 18.62s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 220, Loss: 0.0243, Precision@20: 0.0538, Recall@20: 0.1848\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 23%|██▎       | 231/1000 [1:11:49<3:58:32, 18.61s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 230, Loss: 0.0240, Precision@20: 0.0542, Recall@20: 0.1858\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 24%|██▍       | 241/1000 [1:14:55<3:55:24, 18.61s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 240, Loss: 0.0236, Precision@20: 0.0543, Recall@20: 0.1857\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 25%|██▌       | 251/1000 [1:18:01<3:52:23, 18.62s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 250, Loss: 0.0237, Precision@20: 0.0544, Recall@20: 0.1869\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 26%|██▌       | 261/1000 [1:21:07<3:49:16, 18.62s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 260, Loss: 0.0232, Precision@20: 0.0548, Recall@20: 0.1877\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 27%|██▋       | 271/1000 [1:24:14<3:46:11, 18.62s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 270, Loss: 0.0233, Precision@20: 0.0547, Recall@20: 0.1878\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 28%|██▊       | 281/1000 [1:27:20<3:43:03, 18.61s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 280, Loss: 0.0232, Precision@20: 0.0549, Recall@20: 0.1886\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 29%|██▉       | 291/1000 [1:30:26<3:40:03, 18.62s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 290, Loss: 0.0230, Precision@20: 0.0549, Recall@20: 0.1881\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 30%|███       | 301/1000 [1:33:32<3:36:45, 18.61s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 300, Loss: 0.0228, Precision@20: 0.0550, Recall@20: 0.1893\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 31%|███       | 311/1000 [1:36:38<3:33:34, 18.60s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 310, Loss: 0.0228, Precision@20: 0.0553, Recall@20: 0.1893\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 32%|███▏      | 321/1000 [1:39:44<3:30:36, 18.61s/it]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 320, Loss: 0.0226, Precision@20: 0.0555, Recall@20: 0.1903\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      " 33%|███▎      | 328/1000 [1:41:54<3:28:20, 18.60s/it]"
     ]
    }
   ],
   "source": [
    "best = 0\n",
    "best_model = model.get_embedding(data.edge_index).detach()\n",
    "best_init = model.embedding.weight.detach()\n",
    "p = 0\n",
    "for epoch in tqdm(range(0, 1000)):\n",
    "    loss = train()\n",
    "    precision, recall = test(k=20)\n",
    "    if epoch % 10 == 0:\n",
    "        print(f'Epoch: {epoch:03d}, Loss: {loss:.4f}, Precision@20: '\n",
    "            f'{precision:.4f}, Recall@20: {recall:.4f}')\n",
    "        if recall > best:\n",
    "            p = 0\n",
    "            best_model = model.get_embedding(data.edge_index).detach()\n",
    "            best_init = model.embedding.weight.detach()\n",
    "            best = recall\n",
    "        else:\n",
    "            p += 1\n",
    "            if p > 10:\n",
    "                break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.20"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}