训练tfrecords文件，出现ValueError错误

Question

• Python版本3.6.4，tensorflow版本1.6.0；
• 我自己直接用tensorflow的dataset读取csv文件拿来训练的话，不出现问题；
• 但是我尝试着将csv文件制作成tfrecords文件，然后读入tfrecords文件的话，会出现ValueError: features should be a dictionary of Tensors. Given type: <class 'tensorflow.python.framework.ops.Tensor'>错误；
• 我谷歌了一整天，很多人说是tensorflow版本问题，但是我已经升级到了最新的版本。

所以现在请各位帮忙看看是不是有以下问题：

1. csv文件制作成tfrecords文件错误；
2. 读取制作完的tfrecords文件错误；

运行错误提示如下：

INFO:tensorflow:Using default config.
INFO:tensorflow:Using config: {'_model_dir': 'iris_model_2', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': None, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_service': None, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x120f15eb8>, '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}
INFO:tensorflow:Calling model_fn.
Traceback (most recent call last):
  File "/Users/huanghelin/Desktop/TFrecord/try2.py", line 45, in <module>
    classifier.train(input_fn=lambda: my_input_fn(is_shuffle=True, repeat_count=100))
  File "/Users/huanghelin/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 352, in train
    loss = self._train_model(input_fn, hooks, saving_listeners)
  File "/Users/huanghelin/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 812, in _train_model
    features, labels, model_fn_lib.ModeKeys.TRAIN, self.config)
  File "/Users/huanghelin/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/estimator.py", line 793, in _call_model_fn
    model_fn_results = self._model_fn(features=features, **kwargs)
  File "/Users/huanghelin/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/canned/dnn.py", line 354, in _model_fn
    config=config)
  File "/Users/huanghelin/anaconda3/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/estimator/canned/dnn.py", line 161, in _dnn_model_fn
    'Given type: {}'.format(type(features)))
ValueError: features should be a dictionary of `Tensor`s. Given type: <class 'tensorflow.python.framework.ops.Tensor'>

制作代码代码如下：

import tensorflow as tf
import pandas as pd

tf_name = 'csv.tfrecords'
csv = pd.read_csv("iris_test.csv").values

def _int64_feature(value):
    return tf.train.Feature(int64_list=tf.train.Int64List(value=[value]))

def _float_feature(value):
    # 本来传入的就是个list，所以value不用加[]了
    return tf.train.Feature(float_list=tf.train.FloatList(value=value))

with tf.python_io.TFRecordWriter(tf_name) as writer:
    for row in csv:
        features, label = row[:-1], int(row[-1])
        features = [float(f) for f in features]
        example = tf.train.Example()
        example = tf.train.Example(
            features=tf.train.Features(
                feature={
                    'label': _int64_feature(label),
                    'features': _float_feature(features)
                }
            )
        )
        writer.write(example.SerializeToString())
    print('转化%s完成' % tf_name)

拿去训练的代码：

import tensorflow as tf

tf.logging.set_verbosity(tf.logging.INFO)

feature_names = [
    'SepalLength',
    'SepalWidth',
    'PetalLength',
    'PetalWidth'
]

def my_input_fn(is_shuffle=False, repeat_count=1):
    dataset = tf.data.TFRecordDataset(['csv.tfrecords'])  # filename得是个list

    def parser(record):
        keys_to_features = {
            'label': tf.FixedLenFeature((), dtype=tf.int64),
            'features': tf.FixedLenFeature(shape=(4,), dtype=tf.float32),
        }
        parsed = tf.parse_single_example(record, keys_to_features)
        return parsed['features'], parsed['label']

    dataset = dataset.map(parser)
    # 打乱顺序，buffer_size设置成一个大于数据集中样本数量的值来确保其充分打乱
    if is_shuffle:
        # Randomizes input using a window of 256 elements (read into memory)
        dataset = dataset.shuffle(buffer_size=256)
    # 打包成多少个进行输出
    dataset = dataset.batch(32)
    # 指定要遍历几遍整个数据集
    dataset = dataset.repeat(repeat_count)
    iterator = dataset.make_one_shot_iterator()
    features, labels = iterator.get_next()
    return features, labels

feature_columns = [tf.feature_column.numeric_column(k) for k in feature_names]

classifier = tf.estimator.DNNClassifier(
    feature_columns=feature_columns,  # The input features to our model
    hidden_units=[10, 10],  # Two layers, each with 10 neurons
    n_classes=3,
    model_dir='iris_model_2')  # Path to where checkpoints etc are stored

classifier.train(input_fn=lambda: my_input_fn(is_shuffle=True, repeat_count=100))