stateful LSTM的特點是，在處理過一個batch的訓練數(shù)據(jù)后，其內(nèi)部狀態(tài)（記憶）會被作為下一個batch的訓練數(shù)據(jù)的初始狀態(tài)。狀態(tài)LSTM使得我們可以在合理的計算復雜度內(nèi)處理較長序列
請FAQ中關于stateful LSTM的部分獲取更多信息

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

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

# Expected input batch shape: (batch_size, timesteps, data_dim)
# Note that we have to provide the full batch_input_shape since the network is stateful.
# the sample of index i in batch k is the follow-up for the sample i in batch k-1.
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(10, activation='softmax'))

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

# 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))

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