keras模型第二課 應用舉例 Getting started with the Keras Sequential model - Examples

Getting started with the Keras Sequential model - Examples

翻譯原文：https://keras.io/getting-started/sequential-model-guide/#examples

此鏈接examples folder將會提供一些應用例子，在這些例子中涉及到如下的真實數據集dataset：
· CIPAR10 少量圖像分類：CNN模型
· IMDB 電影評論情感分類：LSTM模型
· 路透社新聞專線主題分類：MLP多層感知器模型
· MNIST 手寫數字分類：MLP & CNN
· 字符級的文本生成：LSTM
等等

多層感知器MLP用於多分類任務

import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
from keras.optimizers import SGD

# 產生numpy數據
import numpy as np
x_train = np.random.random((1000, 20))
y_train = np.random.randint(10, size=(1000, 1))
y_one_hot_train = keras.utils.to_categorical(y_train, num_classes=10)
x_test = np.random.random((100, 20))
y_test = np.random.randint(10, size=(100, 1))
y_one_hot_test = keras.utils.to_categorical(y_test, num_classes=10)

# 構建全連接網絡模型
model = Sequential()
model.add(Dense(64, input_shape=(20, ), activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))

# 編譯模型
sgd = SGD(lr=0.01, momentum=0.9, decay=1e-6, nesterov=True)
model.compile(optimizer=sgd,
              loss=keras.losses.categorical_crossentropy,
              metrics=['accuracy'])

# 訓練模型
model.fit(x_train, y_one_hot_train, epochs=20, batch_size=128)

# 驗證模型
test_score = model.evaluate(x_test, y_one_hot_test, batch_size=128)
print("test_acc: ", test_score[1])

MLP用於二分類任務

import numpy as np
from keras.layers import Dense, Dropout
from keras.models import Sequential

# 產生numpy數據
x_train = np.random.random((1000, 20))
y_train = np.random.randint(2, size=(1000, 1))
x_test = np.random.random((100, 20))
y_test = np.random.randint(2, size=(100, 1))

# 構建模型
model = Sequential()
model.add(Dense(64, input_dim=20, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(64, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

# 編譯模型
model.compile(optimizer=keras.optimizers.Adam(), 
              loss=keras.losses.binary_crossentropy,
              metrics=['accuracy'])

# 訓練模型
model.fit(x_train, y_train, epochs=20, batch_size=128)

# 驗證模型
(test_loss, test_acc) = model.evaluate(x_test, y_test)
print('test_acc: ', test_acc)

類似於VGG的卷積網絡

import numpy as np
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras.optimizers import Adam

# 產生numpy數據
x_train = np.random.random((100, 100, 100, 3))
y_train = np.random.randint(10, size=(100, 1))
y_one_hot_train = keras.utils.to_categorical(y_train, num_classes=10)
x_test = np.random.random((20, 100, 100, 3))
y_test = np.random.randint(10, size=(20, 1))
y_one_hot_test = keras.utils.to_categorical(y_test, num_classes=10)

# 構建模型
model = Sequential()
model.add(Conv2D(32, (3,3), input_shape=(100, 100, 3), activation='relu'))
model.add(Conv2D(32, (3,3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.25))

model.add(Conv2D(64, (3,3), activation='relu'))
model.add(Conv2D(64, (3,3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2,2)))
model.add(Dropout(0.25))

model.add(Flatten())
model.add(Dense(256, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(10, activation='softmax'))

adam = Adam(lr=0.01, decay=1e-6)
model.compile(optimizer=adam,
              loss=keras.losses.categorical_crossentropy,
              metrics=['accuracy'])

model.fit(x_train, y_one_hot_train, epochs=10, batch_size=32)
(test_loss, test_acc) = model.evaluate(x_test, y_one_hot_test)
print('test_acc: ', test_acc)

LSTM用於序列分類: many to one

from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.layers import Embedding, LSTM
from keras.engine.input_layer import Input

# 詞嵌入的詞彙數
vocab_len = 1024
# 詞嵌入的特徵數
emb_dim = 256

# 創建numpy數據 - 模擬文本情感識別數據
max_len = 10 #最長句子長度
x_train = np.random.randint(1000, size=(100, max_len))
y_train = np.random.randint(2, size=(100, 1))
x_test = np.random.randint(1000, size=(10, max_len))
y_test = np.random.randint(2, size=(10, 1))

# 構建模型
model = Sequential()
model.add(Embedding(vocab_len, emb_dim, input_length=max_len))
model.add(LSTM(128))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

# 編譯模型
model.compile(optimizer=keras.optimizers.Adam(),
              loss=keras.losses.binary_crossentropy, 
              metrics=['accuracy'])

# 訓練模型
model.fit(x_train, y_train, batch_size=16, epochs=10)

# 驗證模型
test_loss, test_acc = model.evaluate(x_test, y_test)
print(test_acc)

# 預測模型
x_pred = np.random.randint(1000, size=(1, max_len))
y_pred = model.predict(x_pred)
if y_pred >= 0.5:
    y_pred = 1
else:
    y_pred = 0
print('y_pred: ', y_pred)

不用embedding

from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.layers import Embedding, LSTM
from keras.engine.input_layer import Input

# 序列長度
seq_length = 64
# 通道數
n_channels = 100

# 創造numpy序列數據
x_train = np.random.random((100, seq_length, n_channels))
y_train = np.random.randint(2, size=(100, 1))
x_test = np.random.random((10, seq_length, n_channels))
y_test = np.random.randint(2, size=(10, 1))

# 構建模型
model = Sequential()
model.add(LSTM(256, input_shape=(seq_length, n_channels)))
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

# 編譯模型
model.compile(optimizer=keras.optimizers.Adam(),
              loss=keras.losses.binary_crossentropy, 
              metrics=['accuracy'])

# 訓練模型
model.fit(x_train, y_train, batch_size=16, epochs=10)

# 驗證模型
test_loss, test_acc = model.evaluate(x_test, y_test)
print(test_acc)

# 預測模型
x_pred = x_train = np.random.random((1, seq_length, n_channels))
y_pred = model.predict(x_pred)
if y_pred >= 0.5:
    y_pred = 1
else:
    y_pred = 0
print('y_pred: ', y_pred)

1D卷積網絡用於序列分類：many to one

from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.layers import Embedding
from keras.layers import Conv1D, GlobalAveragePooling1D, MaxPooling1D

# 序列長度
seq_length = 64
# 通道數
n_channels = 100

# 創造numpy序列數據
x_train = np.random.random((100, seq_length, n_channels))
y_train = np.random.randint(2, size=(100, 1))
x_test = np.random.random((10, seq_length, n_channels))
y_test = np.random.randint(2, size=(10, 1))

# 構建模型
model = Sequential()
model.add(Conv1D(64, 3, activation='relu', input_shape=(seq_length, n_channels)))
model.add(Conv1D(64, 3, activation='relu'))
model.add(MaxPooling1D(3))
model.add(Conv1D(128, 3, activation='relu'))
model.add(Conv1D(128, 3, activation='relu'))
model.add(GlobalAveragePooling1D())
model.add(Dropout(0.5))
model.add(Dense(1, activation='sigmoid'))

# 編譯模型
model.compile(optimizer=keras.optimizers.Adam(),
              loss=keras.losses.binary_crossentropy,
              metrics=['accuracy'])

# 訓練模型
model.fit(x_train, y_train, epochs=100, batch_size=16)

# 驗證模型
(test_lose, test_acc) = model.evaluate(x_test, y_test)
print(test_acc)

堆疊的LSTM用於序列分類 many to one

圖中共堆疊了三層LSTM層，前兩層LSTM的輸出爲全序列（return_sequences=True），最後一層LSTM只在最後一個時間步輸出結果（默認）

from keras.models import Sequential
from keras.layers import Dense, LSTM

data_dim = 16 #特徵數
timesteps = 8 #時間步數
num_classes = 10 #分類數

# 創造numpy數據
x_train = np.random.random((100, timesteps, data_dim))
y_train = np.random.randint(num_classes, size=(100, 1))
y_one_hot_train = keras.utils.to_categorical(y_train, num_classes=num_classes)
x_test = np.random.random((10, timesteps, data_dim))
y_test = np.random.randint(num_classes, size=(10, 1))
y_one_hot_test = keras.utils.to_categorical(y_test, num_classes=num_classes)

# 構建模型
model = Sequential()
model.add(LSTM(32, return_sequences=True,
               input_shape=(timesteps, data_dim)))
model.add(LSTM(32, return_sequences=True))
model.add(LSTM(32))
model.add(Dense(num_classes, activation='softmax'))

# 編譯模型
model.compile(optimizer=keras.optimizers.Adam(),
              loss=keras.losses.categorical_crossentropy,
              metrics=['accuracy'])

# 訓練並驗證模型
model.fit(x_train, y_one_hot_train, epochs=100, batch_size=32,
          validation_data=(x_test, y_one_hot_test))

有狀態的LSTM模型用於序列分類 many to one

有狀態的LSTM模型與無狀態的LSTM模型的區別可查看以下博客：Keras 之 LSTM 有狀態模型（stateful LSTM）和無狀態模型（stateless LSTM）

from keras.models import Sequential
from keras.layers import LSTM, Dense

data_dim = 16
timesteps = 8
num_classes = 10
batch_size = 32

# Generate dummy training data
x_train = np.random.random((batch_size * 10, timesteps, data_dim))
y_train = np.random.random((batch_size * 10, num_classes))

# Generate dummy validation data
x_val = np.random.random((batch_size * 3, timesteps, data_dim))
y_val = np.random.random((batch_size * 3, num_classes))

# 構建模型
model = Sequential()
model.add(LSTM(32, return_sequences=True, stateful=True,
               batch_input_shape=(batch_size, timesteps, data_dim)))
model.add(LSTM(32, return_sequences=True, stateful=True))
model.add(LSTM(32, stateful=True))
model.add(Dense(num_classes, activation='softmax'))

# 編譯模型
model.compile(optimizer=keras.optimizers.Adam(),
              loss=keras.losses.categorical_crossentropy,
              metrics=['accuracy'])

# 訓練模型
model.fit(x_train, y_train, 
          batch_size=batch_size, epochs=5, shuffle=False,
          validation_data=(x_val, y_val))