0806 영화분류하기 - 과대적합, 과소적합 (GPU Tensorflow)

레이어가 늘어날수록 웨이팅계수가 많아지기때문에 시간이 더 오래걸림

## 3-1

from keras.datasets import imdb

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

## 3-2

import numpy as np

def vectorize_sequences(sequences, dimension=10000):
    # 크기가 (len(sequences), dimension)) 이고 모든 원소가 0인 행렬을 만듦
    results = np.zeros((len(sequences), dimension))
    for i, sequence in enumerate(sequences) :
        results[i, sequence] = 1. # result[i]에서 특정 인덱스의 위치를 1로 만듦
    return results

# 훈련 데이터를 벡터로 변환
x_train = vectorize_sequences(train_data)
# 테스트 데이터를 벡터로 변환
x_test = vectorize_sequences(test_data)

y_train = np.asarray(train_labels).astype('float32')
y_test = np.asarray(test_labels).astype('float32')​

## 3-3 모델 정의하기

from keras import models
from keras import layers

model = models.Sequential()
Smallermodel = models.Sequential()
Biggermodel = models.Sequential()

## 3-3
model.add(layers.Dense(16, activation='relu', input_shape=(10000,)))
model.add(layers.Dense(16, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))

## 4-4
Smallermodel.add(layers.Dense(6, activation='relu', input_shape=(10000,))) ## 16에서 6으로 축소됨!
Smallermodel.add(layers.Dense(6, activation='relu'))
Smallermodel.add(layers.Dense(1, activation='sigmoid'))

## 4-5
Biggermodel.add(layers.Dense(1024, activation='relu', input_shape=(10000,))) ## 16에서 6으로 축소됨!
Biggermodel.add(layers.Dense(1024, activation='relu'))
Biggermodel.add(layers.Dense(1, activation='sigmoid'))

## 3-4 모델 컴파일 하기

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

## overfitting 을 막기위해 축소된 모델

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

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

## 3-5 옵티마이저 설정하기

from keras import optimizers

model.compile(optimizer = optimizers.RMSprop(lr=0.001),
             loss = 'binary_crossentropy',
             metrics=['accuracy'])

## 3-6 손실과 측정을 함수 객체로 지정하기

from keras import losses
from keras import metrics

model.compile(optimizer=optimizers.RMSprop(lr=0.001),
             loss = losses.binary_crossentropy,
             metrics=[metrics.binary_accuracy])

## 3-7 검증 세트 준비하기

x_val = x_train[:10000]
partial_x_train = x_train[10000:]

y_val=y_train[:10000]
partial_y_train = y_train[10000:]

## 3-8 모델 훈련하기

#model.compile(optimizer='rmsprop',
#             loss = 'binary_crossentropy',
#             metrics=['acc'])

history = model.fit(partial_x_train,
                   partial_y_train,
                   epochs=20,
                   batch_size=512,
                   validation_data = (x_val, y_val))
                  

SmallerHistory = Smallermodel.fit(partial_x_train,
                   partial_y_train,
                   epochs=20,
                   batch_size=512,
                   validation_data = (x_val, y_val))

BiggerHistory = Biggermodel.fit(partial_x_train,
                   partial_y_train,
                   epochs=20,
                   batch_size=512,
                   validation_data = (x_val, y_val))

## 3-9 훈련과 검증손실 그리기

import matplotlib.pyplot as plt

history_dict = history.history
history_dict2 = SmallerHistory.history
history_dict3 = BiggerHistory.history
#loss = history_dict['loss']
val_loss = history_dict['val_loss']
val_loss_small = history_dict2['val_loss']
val_loss_big = history_dict3['val_loss']

epochs = range(1, len(loss) + 1)

plt.plot(epochs, val_loss, '+', label = 'Original model') ## 'bo'는 파란색 점을 의미함
plt.plot(epochs, val_loss_small, 'bo', label = 'Small model')
plt.plot(epochs, val_loss_big, '*', label = 'Big model')
plt.title('Comparison Original,Bigger,Smaller')
plt.xlabel('Epochs')
plt.ylabel('Validation loss')
plt.legend()

plt.show()