所以我的任務(wù)是預(yù)測(cè)序列。我在時(shí)間 t 有 x,y,z 值，它們是浮點(diǎn)型。我必須預(yù)測(cè)在時(shí)間 (t + 1) 具有值 x,y,z 的序列。TIME_STEP = 10N_FEATURES = N_CLASSES = 3LEARNING_RATE = 0.01EPOCHS = 50BATCH_SIZE = 10x = tf.placeholder(tf.float32, shape = [None, N_FEATURES], name = 'name')y = tf.placeholder(tf.float32, shape = [N_CLASSES], name = 'labels')然后我有我的 lstm 模型，它看起來(lái)像：    x = tf.transpose(x, [1, 0])    x = tf.reshape(x, [-1, num_features])    hidden = tf.nn.relu(tf.matmul(x, self.h_W) + self.h_biases)    hidden = tf.split(hidden, self.time_step)    lstm_layers = [tf.contrib.rnn.BasicLSTMCell(self.hidden_units, forget_bias=1.0) for _ in range(2)]    lstm_layers = tf.contrib.rnn.MultiRNNCell(lstm_layers)    outputs, _ = tf.contrib.rnn.static_rnn(lstm_layers, hidden, dtype = tf.float32)    lstm_output = outputs[-1]最后我定義了損失函數(shù)和優(yōu)化器loss = tf.reduce_mean(tf.square(y - y_pred))opt = tf.train.AdamOptimizer(learning_rate = LEARNING_RATE).minimize(loss)現(xiàn)在我想用前 10 個(gè)值來(lái)預(yù)測(cè)第 11 個(gè)值。所以我運(yùn)行會(huì)話for time in range(0, len(X)):        sess.run(opt, feed_dict = {x : X[time: time + TIME_STEP ],                                   y : Y[time + TIME_STEP + 1]})但是當(dāng)我檢查這個(gè)函數(shù)的損失時(shí)，它有很大的價(jià)值，比如 99400290.0，它會(huì)隨著時(shí)間的推移而增加。這是我第一次預(yù)測(cè)序列，所以我想我一定遺漏了一些重要的東西