간단한 자연어 처리 모델

In [1]:

from IPython.core.display import display, HTML

display(HTML("<style>.container {width:90% !important;}</style>"))

필요한 라이브러리 임포트¶

In [2]:

import tensorflow as tf
from tensorflow.keras import preprocessing, layers

입력 및 라벨 데이터 생성¶

In [3]:

samples = ["너 오늘 이뻐 보인다",
           "나는 오늘 기분이 더러워",
           "끝내주는데, 좋은 일이 있나봐",
           "나 좋은 일이 생겼어",
           "아 오늘 진짜 짜증나",
           "환상적인데, 정말 좋은거 같아"]

labels = [[1], [0], [1], [1], [0], [1]]

전처리¶

In [4]:

tokenizer = preprocessing.text.Tokenizer()
tokenizer.fit_on_texts(samples)

word_index = tokenizer.word_index

word_index

Out[4]:

{'오늘': 1,
 '좋은': 2,
 '일이': 3,
 '너': 4,
 '이뻐': 5,
 '보인다': 6,
 '나는': 7,
 '기분이': 8,
 '더러워': 9,
 '끝내주는데': 10,
 '있나봐': 11,
 '나': 12,
 '생겼어': 13,
 '아': 14,
 '진짜': 15,
 '짜증나': 16,
 '환상적인데': 17,
 '정말': 18,
 '좋은거': 19,
 '같아': 20}

In [5]:

sequences = tokenizer.texts_to_sequences(samples)

sequences

Out[5]:

[[4, 1, 5, 6],
 [7, 1, 8, 9],
 [10, 2, 3, 11],
 [12, 2, 3, 13],
 [14, 1, 15, 16],
 [17, 18, 19, 20]]

하이퍼파라미터 설정¶

In [6]:

batch_size = 2
num_epochs = 10
vocab_size = len(word_index) + 1
emb_size = 128
hidden_dimension = 256
output_dimension = 1

심층 신경망 모델 생성¶

In [7]:

class CustomModel(tf.keras.Model):
    
    def __init__(self, vocab_size, embed_dimension, hidden_dimension, output_dimension):
        super(CustomModel, self).__init__(name = "my_model")
        
        self.embedding = layers.Embedding(vocab_size, embed_dimension)
        self.dense_layer = layers.Dense(hidden_dimension, activation = "relu")
        self.output_layer = layers.Dense(output_dimension, activation = "sigmoid")
        
    def call(self, inputs):
        x = self.embedding(inputs)
        x = tf.reduce_mean(x, axis = 1)
        x = self.dense_layer(x)
        x = self.output_layer(x)
        
        return x

모델 훈련¶

In [8]:

model = CustomModel(vocab_size = vocab_size, embed_dimension = emb_size, hidden_dimension = hidden_dimension,\
                    output_dimension = output_dimension)

model.compile(optimizer = tf.keras.optimizers.Adam(0.001), loss = "binary_crossentropy", metrics = ["accuracy"])
model.fit(sequences, labels, epochs = num_epochs, batch_size = batch_size)

Epoch 1/10
3/3 [==============================] - 0s 1ms/step - loss: 0.6914 - accuracy: 0.6667
Epoch 2/10
3/3 [==============================] - 0s 1ms/step - loss: 0.6727 - accuracy: 1.0000
Epoch 3/10
3/3 [==============================] - 0s 1ms/step - loss: 0.6571 - accuracy: 1.0000
Epoch 4/10
3/3 [==============================] - 0s 2ms/step - loss: 0.6425 - accuracy: 1.0000
Epoch 5/10
3/3 [==============================] - 0s 1ms/step - loss: 0.6213 - accuracy: 1.0000
Epoch 6/10
3/3 [==============================] - 0s 1ms/step - loss: 0.6000 - accuracy: 1.0000
Epoch 7/10
3/3 [==============================] - 0s 1ms/step - loss: 0.5724 - accuracy: 1.0000
Epoch 8/10
3/3 [==============================] - 0s 997us/step - loss: 0.5429 - accuracy: 1.0000
Epoch 9/10
3/3 [==============================] - 0s 1ms/step - loss: 0.5068 - accuracy: 1.0000
Epoch 10/10
3/3 [==============================] - 0s 1ms/step - loss: 0.4676 - accuracy: 1.0000

Out[8]:

<tensorflow.python.keras.callbacks.History at 0x200869bf388>