IMDB 데이터를 이용한 간단한 LSTM, GRU, Simple RNN 구현

순환신경망 구현 및 학습¶

In [1]:

import tensorflow as tf

하이퍼 파라미터¶

In [2]:

EPOCHS = 10 # 10번 수행 
NUM_WORDS = 10000

모델 정의¶

In [3]:

class MyModel(tf.keras.Model) :
    def __init__(self):
        super().__init__()
        self.emb = tf.keras.layers.Embedding(NUM_WORDS, 16) 
        # self.lstm = tf.keras.layers.LSTM(32) # 뉴런의 개수
        self.gru = tf.keras.layers.GRU(32) # LSTM의 간소한 버전
        # self.basic = tf.keras.layers.SimpleRNN(32) # LSTM의 간소한 버전
        self.dense = tf.keras.layers.Dense(1, activation = 'sigmoid') # 이진분류 이기때문에 sigmoid
        
    def call(self, x, training = None, mask = None):
        x = self.emb(x)
        #x = self.lstm(x)
        x = self.gru(x)
        #x = self.basic(x)
        return self.dense(x)

데이터셋 준비¶

In [4]:

imdb = tf.keras.datasets.imdb

(x_train, y_train), (x_test, y_test) = imdb.load_data(num_words = NUM_WORDS)

x_train = tf.keras.preprocessing.sequence.pad_sequences(x_train,
                                                        value = 0,
                                                        padding = 'pre',
                                                        maxlen = 32)
x_test = tf.keras.preprocessing.sequence.pad_sequences(x_test,
                                                       value= 0,
                                                       padding = 'pre',
                                                       maxlen = 32 )
train_ds = tf.data.Dataset.from_tensor_slices((x_train, y_train)).shuffle(1000).batch(32)
test_ds = tf.data.Dataset.from_tensor_slices((x_test, y_test)).batch(32)

/home/weaver/anaconda3/envs/study/lib/python3.7/site-packages/tensorflow_core/python/keras/datasets/imdb.py:129: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray
  x_train, y_train = np.array(xs[:idx]), np.array(labels[:idx])
/home/weaver/anaconda3/envs/study/lib/python3.7/site-packages/tensorflow_core/python/keras/datasets/imdb.py:130: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray
  x_test, y_test = np.array(xs[idx:]), np.array(labels[idx:])

모델 생성¶

In [5]:

model = MyModel()
model.compile(optimizer='adam',
              loss='binary_crossentropy', # 긍정, 부정
              metrics=['accuracy'])

학습 루프 동작¶

In [6]:

model.fit(train_ds, validation_data=test_ds, epochs=EPOCHS)

Train for 782 steps, validate for 782 steps
Epoch 1/10
782/782 [==============================] - 12s 15ms/step - loss: 0.5086 - accuracy: 0.7356 - val_loss: 0.4498 - val_accuracy: 0.7881
Epoch 2/10
782/782 [==============================] - 10s 13ms/step - loss: 0.3752 - accuracy: 0.8332 - val_loss: 0.4521 - val_accuracy: 0.7848
Epoch 3/10
782/782 [==============================] - 9s 12ms/step - loss: 0.3312 - accuracy: 0.8597 - val_loss: 0.4877 - val_accuracy: 0.7776
Epoch 4/10
782/782 [==============================] - 10s 12ms/step - loss: 0.2994 - accuracy: 0.8777 - val_loss: 0.5029 - val_accuracy: 0.7752
Epoch 5/10
782/782 [==============================] - 10s 13ms/step - loss: 0.2679 - accuracy: 0.8929 - val_loss: 0.5687 - val_accuracy: 0.7684
Epoch 6/10
782/782 [==============================] - 10s 13ms/step - loss: 0.2275 - accuracy: 0.9141 - val_loss: 0.5958 - val_accuracy: 0.7670
Epoch 7/10
782/782 [==============================] - 10s 13ms/step - loss: 0.1889 - accuracy: 0.9331 - val_loss: 0.7918 - val_accuracy: 0.7542
Epoch 8/10
782/782 [==============================] - 9s 12ms/step - loss: 0.1558 - accuracy: 0.9472 - val_loss: 0.8102 - val_accuracy: 0.7560
Epoch 9/10
782/782 [==============================] - 9s 12ms/step - loss: 0.1265 - accuracy: 0.9587 - val_loss: 0.8782 - val_accuracy: 0.7549
Epoch 10/10
782/782 [==============================] - 9s 12ms/step - loss: 0.1052 - accuracy: 0.9658 - val_loss: 0.9978 - val_accuracy: 0.7470

Out[6]:

<tensorflow.python.keras.callbacks.History at 0x7fecd42990d0>