LSTM 神经网络输入输出层

今天终于弄明白，TensorFlow和Keras中LSTM神经网络的输入输出层到底应该怎么设置和连接了。写个备忘。

https://machinelearningmastery.com/how-to-develop-lstm-models-for-time-series-forecasting/

Stacked LSTM

Multiple hidden LSTM layers can be stacked one on top of another in what is referred to as a Stacked LSTM model.

An LSTM layer requires a three-dimensional input and LSTMs by default will produce a two-dimensional output as an interpretation from the end of the sequence.

We can address this by having the LSTM output a value for each time step in the input data by setting the return_sequences=True argument on the layer. This allows us to have 3D output from hidden LSTM layer as input to the next.

We can, therefore, define a Stacked LSTM as follows.

# define model
model = Sequential()
model.add(LSTM(50, activation='relu', return_sequences=True, input_shape=(n_steps, n_features)))
model.add(LSTM(50, activation='relu'))
model.add(Dense(1))
model.compile(optimizer='adam', loss='mse')

X_train.shape
(500, 40, 1)
y_train.shape
(500, 40, 1)

from keras.models import Sequential
from keras import layers
from keras.optimizers import RMSprop

model = Sequential()
model.add(layers.GRU(100, input_shape=(None, X_train.shape[-1]), return_sequences=True))
model.add(layers.Dense(1))
model.compile(optimizer=RMSprop(), loss='mae')
history = model.fit(X_train, y_train,steps_per_epoch=25,epochs=20)

reset_graph()

n_steps = 40
n_inputs = 1
n_neurons = 100

X = tf.placeholder(tf.float32, [None, n_steps, n_inputs])
y = tf.placeholder(tf.float32, [None, n_steps, n_outputs])

num_units = [500, 200, 100]
cells = [tf.nn.rnn_cell.GRUCell(num_units=n) for n in num_units]
stacked_rnn_cell = tf.nn.rnn_cell.MultiRNNCell(cells)
rnn_outputs, states = tf.nn.dynamic_rnn(stacked_rnn_cell, X, dtype=tf.float32)

# 先去掉一个维度，用一个Dense层连上，再把n_steps这个维度加回去
# [batch_size, n_steps, n_neurons]
# [batch_size * n_steps, n_neurons]
# [batch_size, n_steps, n_neurons]

stacked_rnn_outputs = tf.reshape(rnn_outputs, [-1, n_neurons])
stacked_outputs = tf.layers.dense(stacked_rnn_outputs, n_outputs)
outputs = tf.reshape(stacked_outputs, [-1, n_steps, n_outputs])

loss = tf.reduce_mean(tf.square(outputs - y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate)
training_op = optimizer.minimize(loss)

init = tf.global_variables_initializer()
saver = tf.train.Saver()

n_iterations = 5000
batch_size = 100

with tf.Session() as sess:
    init.run()
    for iteration in range(n_iterations):
        X_batch, y_batch = next_batch(batch_size, n_steps)
        sess.run(training_op, feed_dict={X: X_batch, y: y_batch})
        if iteration % 100 == 0:
            mse = loss.eval(feed_dict={X: X_batch, y: y_batch})
            print(iteration, "\tMSE:", mse)

    X_new = time_series(np.array(t_instance[:-1].reshape(-1, n_steps, n_inputs)))
    y_pred = sess.run(outputs, feed_dict={X: X_new})

    saver.save(sess, "./my_time_series_model")

• 与TensorFlow不同， Keras 中 LSTM 层默认只输出最后一个时间步