我需要開(kāi)發(fā)一個(gè) RNN 模型，并希望使用數(shù)據(jù)生成器來(lái)提供訓(xùn)練/評(píng)估循環(huán)。首先，我在從 csv 文件中獲取數(shù)據(jù)時(shí)使用了這個(gè)幫助功能。RECORD_DEFAULTS_TRAIN = [[0], [0.0], [0.0], [0.0], [0.0], [0.0], [0.0]]def decode_csv(line):   parsed_line = tf.decode_csv(line, RECORD_DEFAULTS_TRAIN)   label =  parsed_line[-1]      # label is the last element of the list   del parsed_line[-1]           # delete the last element from the list   del parsed_line[0]            # even delete the first element bcz it is assumed NOT to be a feature   features = tf.stack(parsed_line)  # Stack features so that you can later vectorize forward prop., etc.   return features, label 這是我的數(shù)據(jù)生成器功能：def data_generator(file_path_list, batch_size):  filenames = tf.placeholder(tf.string, shape=[None])  dataset = tf.data.Dataset.from_tensor_slices(filenames)  dataset = dataset.flat_map(lambda filename: tf.data.TextLineDataset(filename).skip(1).map(decode_csv))  dataset = dataset.shuffle(buffer_size=1000)  dataset = dataset.batch(batch_size)  iterator = dataset.make_initializable_iterator()  next_element = iterator.get_next()  with tf.Session() as sess:      while True:          sess.run(iterator.initializer, feed_dict={filenames: file_path_list})          while True:                          try:                batch_data, batch_labels = sess.run(next_element)                # Dimension of the data needs to be: (batch_size, length_of_each_sequence, nr_inputs_in_each_timestep)                # Since the last batch in a epoch can have a different size,                # "batch_data.shape[0]" is used instead of batch_size                batch_data = np.reshape(batch_data, (batch_data.shape[0], SEQUENCE_LEN, 1))              except tf.errors.OutOfRangeError:                break              yield (batch_data, batch_labels)