Merge pull request #12 from recohut/stage

Stage

_notebooks/2022-01-10-fm-ml.ipynb

@@ -0,0 +1 @@
+{"cells":[{"cell_type":"markdown","source":["# FM on ML-100k in PyTorch"],"metadata":{"id":"OXuS4HHuT8wC"}},{"cell_type":"code","source":["!wget -q --show-progress https://files.grouplens.org/datasets/movielens/ml-100k.zip\n","!unzip ml-100k.zip\n","!mv ml-100k/u.data ."],"metadata":{"id":"AJhJHCd6TT-d"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["import numpy as np\n","import pandas as pd\n","import torch\n","from torch import nn\n","from torch.utils.data import Dataset, DataLoader\n","from sklearn.utils import shuffle\n","from sklearn.model_selection import train_test_split\n","from sklearn.metrics import mean_squared_error\n","from tqdm import tqdm\n","\n","learning_rate = 1e-3\n","weight_decay = 1e-4\n","epochs = 10\n","batch_szie = 1024\n","min_val, max_val = 1.0, 5.0\n","device = torch.device('cpu')\n","id_embedding_dim = 256  # id嵌入向量的长度\n","\n","\n","# fm模型\n","class FmLayer(nn.Module):\n","\n","    def __init__(self, p, k):\n","        super(FmLayer, self).__init__()\n","        self.p, self.k = p, k\n","        self.linear = nn.Linear(self.p, 1, bias=True)\n","        self.v = nn.Parameter(torch.Tensor(self.p, self.k), requires_grad=True)\n","        self.v.data.uniform_(-0.01, 0.01)\n","        self.drop = nn.Dropout(0.2)\n","\n","    def forward(self, x):\n","        linear_part = self.linear(x)\n","        inter_part1 = torch.pow(torch.mm(x, self.v), 2)\n","        inter_part2 = torch.mm(torch.pow(x, 2), torch.pow(self.v, 2))\n","        pair_interactions = torch.sum(torch.sub(inter_part1, inter_part2), dim=1)\n","        self.drop(pair_interactions)\n","        output = linear_part.transpose(1, 0) + 0.5 * pair_interactions\n","        return output.view(-1, 1)\n","\n","\n","class FM(nn.Module):\n","    def __init__(self, user_nums, item_nums, id_embedding_dim):\n","        super(FM, self).__init__()\n","        # 对用户， 物品的id进行嵌入\n","        self.user_id_vec = nn.Embedding(user_nums, id_embedding_dim)\n","        self.item_id_vec = nn.Embedding(item_nums, id_embedding_dim)\n","\n","        self.fm = FmLayer(id_embedding_dim * 2, 10)\n","\n","    def forward(self, u_id, i_id):\n","        u_vec = self.user_id_vec(u_id)\n","        i_vec = self.item_id_vec(i_id)\n","        x = torch.cat((u_vec, i_vec), dim=1)\n","        rate = self.fm(x)\n","        return rate\n","\n","\n","class FmDataset(Dataset):\n","    def __init__(self, uid, iid, rating):\n","        self.uid = uid\n","        self.iid = iid\n","        self.rating = rating\n","\n","    def __getitem__(self, index):\n","        return self.uid[index], self.iid[index], self.rating[index]\n","\n","    def __len__(self):\n","        return len(self.uid)\n","\n","\n","def train_iter(model, optimizer, data_loder, criterion):\n","    model.train()\n","    total_loss = 0\n","    total_len = 0\n","\n","    for index, (x_u, x_i, y) in enumerate(data_loder):\n","        x_u, x_i, y = x_u.to(device), x_i.to(device), y.to(device)\n","        y = (y - min_val) / (max_val - min_val) + 0.01\n","        y_pre = model(x_u, x_i)\n","\n","        loss = criterion(y.view(-1, 1), y_pre)\n","        optimizer.zero_grad()\n","        loss.backward()\n","        optimizer.step()\n","\n","        total_loss += loss.item() * len(y_pre)\n","        total_len += len(y_pre)\n","\n","    loss = total_loss / total_len\n","    return loss\n","\n","\n","def val_iter(model, data_loader):\n","    model.eval()\n","    labels, predicts = list(), list()\n","\n","    with torch.no_grad():\n","        for x_u, x_i, y in data_loader:\n","            x_u, x_i, y = x_u.to(device), x_i.to(device), y.to(device)\n","            y_pre = model(x_u, x_i)\n","            y_pre = min_val + (y_pre - 0.01) * (max_val - min_val)\n","            y_pre = torch.where(y_pre > 5.0, torch.full_like(y_pre, 5.0), y_pre)\n","            y_pre = torch.where(y_pre < 1.0, torch.full_like(y_pre, 1.0), y_pre)\n","            labels.extend(y.tolist())\n","            predicts.extend(y_pre.tolist())\n","    mse = mean_squared_error(np.array(labels), np.array(predicts))\n","\n","    return mse\n","\n","\n","def main():\n","    df = pd.read_csv('u.data', header=None, delimiter='\\t')\n","    len_df, u_max_id, i_max_id = len(df), max(df[0]) + 1, max(df[1]) + 1\n","    print(df.shape, max(df[0]), max(df[1]))\n","    x, y = df.iloc[:, :2], df.iloc[:, 2]\n","    x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=2020)\n","    x_train, x_val, y_train, y_val = train_test_split(x_train, y_train, test_size=0.2, random_state=2020)\n","    train_loader = DataLoader(\n","        FmDataset(np.array(x_train[0]), np.array(x_train[1]), np.array(y_train).astype(np.float32)), batch_size=batch_szie)\n","    val_loader = DataLoader(FmDataset(np.array(x_val[0]), np.array(x_val[1]), np.array(y_val).astype(np.float32)), batch_size=batch_szie)\n","    test_loader = DataLoader(FmDataset(np.array(x_test[0]), np.array(x_test[1]), np.array(y_test).astype(np.float32)), batch_size=batch_szie)\n","\n","    # 模型初始化\n","    model = FM(u_max_id, i_max_id, id_embedding_dim).to(device)\n","    optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate, weight_decay=weight_decay)\n","    loss_func = torch.nn.MSELoss().to(device)\n","\n","    # 训练模型\n","    best_val_mse, best_val_epoch = 10, 0\n","    for epoch in range(epochs):\n","        loss = train_iter(model, optimizer, train_loader, loss_func)\n","        mse = val_iter(model, val_loader)\n","        print(\"epoch:{}, loss:{:.5}, mse:{:.5}\".format(epoch, loss, mse))\n","        if best_val_mse > mse:\n","            best_val_mse, best_val_epoch = mse, epoch\n","            torch.save(model, 'best_model')\n","    print(\"best val epoch is {}, mse is {}\".format(best_val_epoch, best_val_mse))\n","    model = torch.load('best_model').to(device)\n","    test_mse = val_iter(model, test_loader)\n","    print(\"test mse is {}\".format(test_mse))\n","\n","\n","if __name__ == '__main__':\n","    main()"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"q8rEahLvTbqF","executionInfo":{"status":"ok","timestamp":1641538334270,"user_tz":-330,"elapsed":12873,"user":{"displayName":"Sparsh Agarwal","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13037694610922482904"}},"outputId":"20358627-dede-48d7-8a5f-88236d55aaae"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["(100000, 4) 943 1682\n","epoch:0, loss:0.2471, mse:1.7855\n","epoch:1, loss:0.095476, mse:1.3192\n","epoch:2, loss:0.074207, mse:1.1511\n","epoch:3, loss:0.066409, mse:1.0806\n","epoch:4, loss:0.062901, mse:1.046\n","epoch:5, loss:0.060974, mse:1.0257\n","epoch:6, loss:0.059704, mse:1.0118\n","epoch:7, loss:0.058753, mse:1.0015\n","epoch:8, loss:0.057982, mse:0.99323\n","epoch:9, loss:0.057325, mse:0.98636\n","best val epoch is 9, mse is 0.9863620211966587\n","test mse is 0.993012835364686\n"]}]},{"cell_type":"markdown","source":["---"],"metadata":{"id":"rsNyAAaUT5lA"}},{"cell_type":"code","source":["!pip install -q watermark\n","%reload_ext watermark\n","%watermark -a \"Sparsh A.\" -m -iv -u -t -d"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"Kk0V69zDT5lE","executionInfo":{"status":"ok","timestamp":1641538388085,"user_tz":-330,"elapsed":4265,"user":{"displayName":"Sparsh Agarwal","photoUrl":"https://lh3.googleusercontent.com/a/default-user=s64","userId":"13037694610922482904"}},"outputId":"5dfb8942-7c65-4d4f-c546-382b6a4f04af"},"execution_count":null,"outputs":[{"output_type":"stream","name":"stdout","text":["Author: Sparsh A.\n","\n","Last updated: 2022-01-07 06:53:08\n","\n","Compiler    : GCC 7.5.0\n","OS          : Linux\n","Release     : 5.4.144+\n","Machine     : x86_64\n","Processor   : x86_64\n","CPU cores   : 2\n","Architecture: 64bit\n","\n","pandas : 1.1.5\n","numpy  : 1.19.5\n","IPython: 5.5.0\n","torch  : 1.10.0+cu111\n","\n"]}]},{"cell_type":"markdown","source":["---"],"metadata":{"id":"Nyd-d0uGT5lG"}},{"cell_type":"markdown","source":["**END**"],"metadata":{"id":"rVlR650LT5lG"}}],"metadata":{"kernelspec":{"display_name":"Python 3","name":"python3"},"colab":{"name":"2022-01-10-fm-ml.ipynb","provenance":[{"file_id":"https://github.com/recohut/nbs/blob/main/raw/P176240%20%7C%20FM%20on%20ML-100k%20in%20PyTorch.ipynb","timestamp":1644598926871},{"file_id":"1Qbw_3372DrLKATKz_66dgUd5EdA_8lwh","timestamp":1641538393107},{"file_id":"1npp4hgFBQRflbqyRW4TayaUFIJJA_xHo","timestamp":1639737398037},{"file_id":"1vh6Mr1C7uh08K4zR4B2VIfqsK22pkraT","timestamp":1639730564985},{"file_id":"1F1wdk7jG5W0jbVM1nZBOPV0TYuSO7frV","timestamp":1639730030880},{"file_id":"https://github.com/RecoHut-Projects/recohut/blob/S394070/nbs/models/tensorflow/deepmf.ipynb","timestamp":1639729505410}],"collapsed_sections":[]}},"nbformat":4,"nbformat_minor":0}