DNN(Deep Neural Network)-IMDB _Binary Classification

NLP(Natural Language Processing)

import tensorflow as tf

 

1. IMDB Data_Set Load & Review

1-1. Load IMDB Data_Set

• Word to Vector
• 전체 데이터 내에서 단어의 사용빈도에 따라 인덱스화
• 정수 인덱스 '11'은 11번째로 자주 사용된 단어를 나타냄
• num_words = 10000 : 인덱스 값 10000 이하의 단어만 추출
• 단어 인덱스 값이 10000을 넘지 않는 단어만 분석에 사용

from tensorflow.keras.datasets import imdb

(train_data, train_labels), (test_data, test_labels) = imdb.load_data(num_words = 10000)

max(max(W) for W in train_data)

9999

---

1-2. Visualization & Frequency(Optional)

import matplotlib.pyplot as plt

print('리뷰 최대 길이 :', max(len(L) for L in train_data))
print('리뷰 평균 길이 :', sum(map(len, train_data))/len(train_data))

plt.figure(figsize = (9, 6))
plt.hist([len(L) for L in train_data], bins = 50)
plt.xlabel('Length of train_data')
plt.ylabel('Number of train_data')
plt.show()

import numpy as np

unique_elements, counts_elements = np.unique(train_labels, return_counts = True)

print('Label 빈도수:')
print(np.asarray((unique_elements, counts_elements)))

---

1-3. Data Structure Review(Optional)

# 전체 train_data 개수
print(len(train_data))

# 첫번째 train_data 정보
print(len(train_data[0]))
print(train_data[0][0:10])
print(train_data[0].count(4))
print(train_labels[0])

---

1-4. Vector to Word(Optional)

word_index = imdb.get_word_index()  #get_word_index( ) : 단어와 인덱스를 매핑한 사전

print(word_index)

# 0, 1, 2 : '패딩', '문서 시작', '사전에 없음'

# 인덱스와 단어 위치 변경

reverse_word_index = dict([(value, key) for (key, value) in word_index.items()])

print(reverse_word_index)

#0번 영화 리뷰 디코딩(1:긍정)

decoded_review = ' '.join([reverse_word_index.get(i - 3, '?') for i in train_data[0]])

print(decoded_review)
print(train_labels[0])

#1번 영화리뷰 디코딩(0:부정)

ecoded_review = ' '.join([reverse_word_index.get(i - 3, '?') for i in train_data[1]])

print(decoded_review)
print(train_labels[1])

---

2. Tensor Transformation

2-1. X_train & X_test : (25000, 10000)

• vectorize_sequence( ) 정의
• 크기는 10000이고 모든 원소가 0인 행렬 생성
  • np.zeros(len(sequences), dimension))
• 값이 존재하는 인덱스의 위치를 1로 지정
  • enumerate( )
  • results[i, sequence] = 1.0

import numpy as np

def vectorize_sequences(sequences, dimension = 10000):
    results = np.zeros((len(sequences), dimension))
    for i, sequence in enumerate(sequences):
        results[i, sequence] = 1.0
    return results

r = np.zeros((5, 10))
v = [1, 3, 5, 7, 9]

for i, v in enumerate(v):
    r[i, v] = 1.0

print(r)

X_train = vectorize_sequences(train_data)
X_test = vectorize_sequences(test_data)

X_train.shape, X_test.shape

((25000, 10000), (25000, 10000))

print(X_train[0][:21])
print(X_train[0][9979:])

print(X_test[0][:21])
print(X_test[0][9979:])

---

2-2. y_train & y_test

y_train = np.asarray(train_labels).astype(float)
y_test = np.asarray(test_labels).astype(float)

print(y_train[:21])
print(y_test[:21])

---

2-3. Train vs. Validation Split

X_valid = X_train[:10000]
partial_X_train = X_train[10000:]

y_valid = y_train[:10000]
partial_y_train = y_train[10000:]

partial_X_train.shape, partial_y_train.shape, X_valid.shape, y_valid.shape

((15000, 10000), (15000,), (10000, 10000), (10000,))

---

3. IMDB Keras Modeling

3-1. Model Define

from tensorflow.keras import models
from tensorflow.keras import layers

imdb = models.Sequential()
imdb.add(layers.Dense(16, activation = 'relu', input_shape = (10000,)))
imdb.add(layers.Dense(16, activation = 'relu'))
imdb.add(layers.Dense(1, activation = 'sigmoid'))

imdb.summary()

---

3-2. Model Compile

imdb.compile(loss = 'binary_crossentropy',
             optimizer = 'rmsprop',
             metrics = ['accuracy'])

---

3-3. Model Fit

%%time

Hist_imdb = imdb.fit(partial_X_train, partial_y_train,
                     epochs = 50,
                     batch_size = 512,
                     validation_data = (X_valid, y_valid))

---

3-4.학습 결과 시각화

import matplotlib.pyplot as plt

epochs = range(1, len(Hist_imdb.history['loss']) + 1)

plt.figure(figsize = (9, 6))
plt.plot(epochs, Hist_imdb.history['loss'])
plt.plot(epochs, Hist_imdb.history['val_loss'])
plt.title('Training & Validation Loss')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend(['Training Loss', 'Validation Loss'])
plt.grid()
plt.show()

import matplotlib.pyplot as plt

epochs = range(1, len(Hist_imdb.history['accuracy']) + 1)

plt.figure(figsize = (9, 6))
plt.plot(epochs, Hist_imdb.history['accuracy'])
plt.plot(epochs, Hist_imdb.history['val_accuracy'])
plt.title('Training & Validation Accuracy')
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.legend(['Training Accuracy', 'Validation Accuracy'])
plt.grid()
plt.show()

---

3-5.Model Evaluate

loss, accuracy = imdb.evaluate(X_test, y_test)

print('Loss = {:.5f}'.format(loss))
print('Accuracy = {:.5f}'.format(accuracy))

---

3-6. Model Predict

np.round(imdb.predict(X_test))