{"id":29939,"date":"2024-10-28T03:18:46","date_gmt":"2024-10-28T03:18:46","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=29939"},"modified":"2024-11-26T06:50:46","modified_gmt":"2024-11-26T06:50:46","slug":"%eb%94%a5%eb%9f%ac%eb%8b%9d-%ed%8c%8c%ec%9d%b4%ed%86%a0%ec%b9%98-%ea%b0%95%ec%a2%8c-rnn-lstm-gru-%ec%84%b1%eb%8a%a5-%eb%b9%84%ea%b5%90","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/29939\/","title":{"rendered":"\ub525\ub7ec\ub2dd \ud30c\uc774\ud1a0\uce58 \uac15\uc88c, RNN, LSTM, GRU \uc131\ub2a5 \ube44\uad50"},"content":{"rendered":"

<\/p>\n

\ub525\ub7ec\ub2dd\uc740 \uc624\ub298\ub0a0 \ub370\uc774\ud130 \uacfc\ud559\uacfc \uc778\uacf5\uc9c0\ub2a5 \ubd84\uc57c\uc5d0\uc11c \ud544\uc218\uc801\uc778 \uae30\uc220\ub85c \uc790\ub9ac\uc7a1\uace0 \uc788\uc2b5\ub2c8\ub2e4.
\n \uc774 \uac15\uc88c\uc5d0\uc11c\ub294 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub294 \uc8fc\uc694 \uc778\uacf5\uc2e0\uacbd\ub9dd \uad6c\uc870\uc778 RNN(\uc21c\ud658 \uc2e0\uacbd\ub9dd), LSTM(\uc7a5\uae30 \ub2e8\uae30 \uae30\uc5b5), GRU(\uac8c\uc774\ud2b8 \uc21c\ud658 \uc720\ub2db)\uc5d0 \ub300\ud574 \uc2ec\ub3c4 \uc788\uac8c \ub17c\uc758\ud558\uace0 \uac01 \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ube44\uad50\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n

1. RNN(\uc21c\ud658 \uc2e0\uacbd\ub9dd) \uc774\ud574\ud558\uae30<\/h2>\n

RNN\uc740 \uc21c\ucc28\uc801\uc73c\ub85c \uc785\ub825\ub418\ub294 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\ub3c4\ub85d \uc124\uacc4\ub41c \uc2e0\uacbd\ub9dd\uc785\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc778 \uc2e0\uacbd\ub9dd\uacfc \ub2ec\ub9ac RNN\uc740 \uc774\uc804\uc758 \ucd9c\ub825 \uc0c1\ud0dc\ub97c \ud604\uc7ac\uc758 \uc785\ub825\uc73c\ub85c \uc0ac\uc6a9\ud558\uc5ec \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\uc758 \uc2dc\uac04\uc801 \uc758\uc874\uc131\uc744 \ud559\uc2b5\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n

1.1. RNN \uad6c\uc870<\/h3>\n

RNN\uc758 \uae30\ubcf8 \uad6c\uc870\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n

\n    h_t = f(W_hh * h_{t-1} + W_xh * x_t)\n    <\/code><\/pre>\n
\uc5ec\uae30\uc11c h_t<\/code>\ub294 \ud604\uc7ac \uc0c1\ud0dc, h_{t-1}<\/code>\ub294 \uc774\uc804 \uc0c1\ud0dc, x_t<\/code>\ub294 \ud604\uc7ac \uc785\ub825, W_hh<\/code>\uc640 W_xh<\/code>\ub294 \uac00\uc911\uce58 \ub9e4\uac1c\ubcc0\uc218, f<\/code>\ub294 \ud65c\uc131\ud654 \ud568\uc218\uc785\ub2c8\ub2e4.<\/p>\n
1.2. RNN\uc758 \ud55c\uacc4<\/h3>\n
RNN\uc740 \uc7a5\uae30 \uc758\uc874\uc131 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ub370 \uc5b4\ub824\uc6c0\uc774 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub294 RNN\uc774 \uae34 \uc2dc\ud000\uc2a4\uc5d0\uc11c \uc624\ub79c \uc2dc\uac04 \uc804\uc5d0 \ubc1c\uc0dd\ud55c \uc815\ubcf4\ub97c \uae30\uc5b5\ud558\uae30 \uc5b4\ub835\uae30 \ub54c\ubb38\uc785\ub2c8\ub2e4.<\/p>\n
2. LSTM(\uc7a5\uae30 \ub2e8\uae30 \uae30\uc5b5)\uc758 \ub3c4\uc785<\/h2>\n
LSTM\uc740 RNN\uc758 \ud55c\uacc4\ub97c \uadf9\ubcf5\ud558\uae30 \uc704\ud574 \uace0\uc548\ub41c \uad6c\uc870\ub85c, \uae34 \uc2dc\ud000\uc2a4 \ub370\uc774\ud130\uc758 \ud559\uc2b5\uc5d0 \uac15\ub825\ud55c \uc131\ub2a5\uc744 \ubc1c\ud718\ud569\ub2c8\ub2e4.<\/p>\n
2.1. LSTM \uad6c\uc870<\/h3>\n
LSTM\uc740 \uc140 \uc0c1\ud0dc(cell state)\uc640 \uac8c\uc774\ud2b8(gate) \uba54\ucee4\ub2c8\uc998\uc744 \ud1b5\ud574 \uc815\ubcf4\ub97c \uc120\ud0dd\uc801\uc73c\ub85c \uae30\uc5b5\ud558\uace0 \uc78a\ub294 \uc5ed\ud65c\uc744 \ud569\ub2c8\ub2e4. LSTM\uc758 \uae30\ubcf8 \uc218\uc2dd\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n
\n    f_t = \u03c3(W_f * [h_{t-1}, x_t] + b_f)  \/\/ Forget gate\n    i_t = \u03c3(W_i * [h_{t-1}, x_t] + b_i)  \/\/ Input gate\n    o_t = \u03c3(W_o * [h_{t-1}, x_t] + b_o)  \/\/ Output gate\n    C_t = f_t * C_{t-1} + i_t * tanh(W_c * [h_{t-1}, x_t] + b_c)  \/\/ Cell state update\n    h_t = o_t * tanh(C_t)  \/\/ Final output\n    <\/code><\/pre>\n
2.2. LSTM\uc758 \uc7a5\uc810<\/h3>\n
LSTM\uc740 \uae34 \uc2dc\ud000\uc2a4\uc5d0\uc11c\ub3c4 \uc815\ubcf4\uc758 \ud750\ub984\uc744 \uc6d0\ud65c\ud558\uac8c \uc720\uc9c0\ud560 \uc218 \uc788\uc73c\uba70, \ub525\ub7ec\ub2dd \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ub192\uc774\ub294 \ub370 \uac15\ub825\ud55c \ub3c4\uad6c\uc785\ub2c8\ub2e4.<\/p>\n
3. GRU(\uac8c\uc774\ud2b8 \uc21c\ud658 \uc720\ub2db)\uc758 \ube44\uad50<\/h2>\n
GRU\ub294 LSTM\uc744 \ub2e8\uc21c\ud654\ud55c \ubaa8\ub378\ub85c, \ub354 \uc801\uc740 \ud30c\ub77c\ubbf8\ud130\ub85c \uc720\uc0ac\ud55c \uc131\ub2a5\uc744 \ubc1c\ud718\ud569\ub2c8\ub2e4.<\/p>\n
3.1. GRU \uad6c\uc870<\/h3>\n
\n    z_t = \u03c3(W_z * [h_{t-1}, x_t] + b_z)  \/\/ Update gate\n    r_t = \u03c3(W_r * [h_{t-1}, x_t] + b_r)  \/\/ Reset gate\n    h_t = (1 - z_t) * h_{t-1} + z_t * tanh(W_h * [r_t * h_{t-1}, x_t] + b_h)  \/\/ Final output\n    <\/code><\/pre>\n
3.2. GRU\uc758 \uc7a5\uc810<\/h3>\n
GRU\ub294 LSTM\ubcf4\ub2e4 \uc131\ub2a5\uc774 \ube44\uc2b7\ud558\uba74\uc11c\ub3c4 \uc801\uc740 \uc790\uc6d0\uc73c\ub85c \ud6c8\ub828\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ub610\ud55c, \uc0c1\ub300\uc801\uc73c\ub85c \ub354 \uac04\ub2e8\ud55c \uad6c\uc870 \ub355\ubd84\uc5d0 \uacc4\uc0b0 \ud6a8\uc728\uc131\uc774 \ub192\uc544\uc9c0\ub294 \uc7a5\uc810\uc774 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
4. RNN, LSTM, GRU \uc131\ub2a5 \ube44\uad50 \uc2e4\uc2b5<\/h2>\n
\uc774\uc81c PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec RNN, LSTM, GRU \ubaa8\ub378\uc744 \uad6c\ud604\ud558\uace0 \uc131\ub2a5\uc744 \ube44\uad50\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uac04\ub2e8\ud55c \uc2dc\uacc4\uc5f4 \uc608\uce21 \ubb38\uc81c\ub85c \uc9c4\ud589\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
4.1. \ub370\uc774\ud130 \uc900\ube44<\/h3>\n
\uc544\ub798 \ucf54\ub4dc\ub294 \uac04\ub2e8\ud55c \uc2dc\uacc4\uc5f4 \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/p>\n
\nimport numpy as np\nimport torch\nimport torch.nn as nn\nimport matplotlib.pyplot as plt\n\n# \uc2dc\uacc4\uc5f4 \ub370\uc774\ud130 \uc0dd\uc131\ndef create_dataset(seq, time_step=1):\n    X, Y = [], []\n    for i in range(len(seq) - time_step - 1):\n        X.append(seq[i:(i + time_step)])\n        Y.append(seq[i + time_step])\n    return np.array(X), np.array(Y)\n\n# \uc2dc\uacc4\uc5f4 \ub370\uc774\ud130\ndata = np.sin(np.arange(0, 100, 0.1))\ntime_step = 10\nX, Y = create_dataset(data, time_step)\n\n# PyTorch \ud150\uc11c \ubcc0\ud658\nX = torch.FloatTensor(X).view(-1, time_step, 1)\nY = torch.FloatTensor(Y)\n    <\/code><\/pre>\n
4.2. \ubaa8\ub378 \uad6c\ud604<\/h3>\n
\uac01 \ubaa8\ub378\uc744 \uad6c\ud604\ud569\ub2c8\ub2e4. RNN \ubaa8\ub378\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n
\nclass RNNModel(nn.Module):\n    def __init__(self, input_size, hidden_size):\n        super(RNNModel, self).__init__()\n        self.rnn = nn.RNN(input_size, hidden_size, batch_first=True)\n        self.fc = nn.Linear(hidden_size, 1)\n\n    def forward(self, x):\n        out, _ = self.rnn(x)\n        out = self.fc(out[:, -1, :])\n        return out\n\n# \ubaa8\ub378 \ucd08\uae30\ud654\nrnn_model = RNNModel(input_size=1, hidden_size=5)\n    <\/code><\/pre>\n
\uc774\uc81c LSTM \ubaa8\ub378\uc744 \uad6c\ud604\ud569\ub2c8\ub2e4:<\/p>\n
\nclass LSTMModel(nn.Module):\n    def __init__(self, input_size, hidden_size):\n        super(LSTMModel, self).__init__()\n        self.lstm = nn.LSTM(input_size, hidden_size, batch_first=True)\n        self.fc = nn.Linear(hidden_size, 1)\n\n    def forward(self, x):\n        out, _ = self.lstm(x)\n        out = self.fc(out[:, -1, :])\n        return out\n\n# \ubaa8\ub378 \ucd08\uae30\ud654\nlstm_model = LSTMModel(input_size=1, hidden_size=5)\n    <\/code><\/pre>\n
\ub9c8\uc9c0\ub9c9\uc73c\ub85c GRU \ubaa8\ub378\uc744 \uad6c\ud604\ud558\uaca0\uc2b5\ub2c8\ub2e4:<\/p>\n
\nclass GRUModel(nn.Module):\n    def __init__(self, input_size, hidden_size):\n        super(GRUModel, self).__init__()\n        self.gru = nn.GRU(input_size, hidden_size, batch_first=True)\n        self.fc = nn.Linear(hidden_size, 1)\n\n    def forward(self, x):\n        out, _ = self.gru(x)\n        out = self.fc(out[:, -1, :])\n        return out\n\n# \ubaa8\ub378 \ucd08\uae30\ud654\ngru_model = GRUModel(input_size=1, hidden_size=5)\n    <\/code><\/pre>\n
4.3. \ubaa8\ub378 \ud6c8\ub828<\/h3>\n
\ubaa8\ub378\uc744 \ud6c8\ub828\uc2dc\ud0a4\uace0 \uc131\ub2a5\uc744 \ube44\uad50\ud569\ub2c8\ub2e4.<\/p>\n
\ndef train_model(model, X_train, Y_train, num_epochs=100, learning_rate=0.01):\n    criterion = nn.MSELoss()\n    optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)\n\n    for epoch in range(num_epochs):\n        model.train()\n        optimizer.zero_grad()\n        outputs = model(X_train)\n        loss = criterion(outputs, Y_train.view(-1, 1))\n        loss.backward()\n        optimizer.step()\n\n    return model\n\n# \ubaa8\ub378 \ud6c8\ub828\nrnn_trained = train_model(rnn_model, X, Y)\nlstm_trained = train_model(lstm_model, X, Y)\ngru_trained = train_model(gru_model, X, Y)\n    <\/code><\/pre>\n
4.4. \uc131\ub2a5 \ud3c9\uac00<\/h3>\n
\uac01 \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud569\ub2c8\ub2e4.<\/p>\n
\ndef evaluate_model(model, X_test):\n    model.eval()\n    with torch.no_grad():\n        predictions = model(X_test)\n    return predictions\n\n# \uc608\uce21\nrnn_predictions = evaluate_model(rnn_trained, X)\nlstm_predictions = evaluate_model(lstm_trained, X)\ngru_predictions = evaluate_model(gru_trained, X)\n\n# \uacb0\uacfc \uc2dc\uac01\ud654\nplt.figure(figsize=(12, 8))\nplt.plot(Y.numpy(), label='True')\nplt.plot(rnn_predictions.numpy(), label='RNN Predictions')\nplt.plot(lstm_predictions.numpy(), label='LSTM Predictions')\nplt.plot(gru_predictions.numpy(), label='GRU Predictions')\nplt.legend()\nplt.show()\n    <\/code><\/pre>\n
5. \uacb0\ub860<\/h2>\n
\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 RNN, LSTM, GRU\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \uac01\uac01\uc758 \uad6c\ud604 \ubc29\ubc95, \uadf8\ub9ac\uace0 \uc131\ub2a5 \ube44\uad50\ub97c \ud1b5\ud574 \uc774\ub4e4 \ubaa8\ub378\uc758 \ud2b9\uc131\uc744 \uc774\ud574\ud588\uc2b5\ub2c8\ub2e4. RNN\uc740 \uac00\uc7a5 \uae30\ubcf8\uc801\uc778 \ud615\ud0dc\ub85c, LSTM\uacfc GRU\ub294 \uac01\uac01\uc758 \ud544\uc694\uc5d0 \ub530\ub77c \uc120\ud0dd\ud558\uc5ec \uc0ac\uc6a9\ud560 \uc218 \uc788\ub294 \uac15\ub825\ud55c \ub3c4\uad6c\ub4e4\uc785\ub2c8\ub2e4. \ube44\uc988\ub2c8\uc2a4 \ubb38\uc81c\uc5d0 \ub530\ub77c \uc801\uc808\ud55c \ubaa8\ub378\uc744 \uc120\ud0dd\ud558\ub294 \uac83\uc774 \uc911\uc694\ud569\ub2c8\ub2e4.<\/p>\n
\ucc38\uace0 \uc790\ub8cc<\/h2>\n
\ucd94\uac00\uc801\uc778 \ud559\uc2b5\uc744 \uc704\ud574 \ub2e4\uc74c \ub9ac\uc18c\uc2a4\ub97c \ucc38\uace0\ud558\uc138\uc694:<\/p>\n