Python 3.x 使用RESTAPI查询tensorflow_model_服务器文本分类器将返回无效参数

我打算运行一个

tensorflow

服务器，该服务器可以使用字符串数组进行查询，并返回每个字符串的分类

我的估计器如下所示：

def _initialise_estimator():

   feature_columns = _create_feature_columns()

   # Set run configs. Only keep one checkpoint.
   config = tensorflow.estimator.RunConfig(
      keep_checkpoint_max = 1,
   )

   # Create the estimator.
   estimator = tensorflow.estimator.DNNClassifier(
      hidden_units = [500, 100],
      feature_columns = feature_columns,
      n_classes = 98,
      optimizer = tensorflow.train.AdagradOptimizer(learning_rate = 0.003),
      model_dir = _model_location,
      config = config
   )

   return estimator


def _create_feature_columns():
   # Feature column. Assumed to be called description.
   return [tensorflow_hub.text_embedding_column(
      key = 'description',
      module_spec = 'https://tfhub.dev/google/nnlm-en-dim128/1'
   )]

_estimator = _initialise_estimator()

tensorflow_model_server --rest_api_port=8501 --model_name=prediction --model_base_path=/app/prediction/lib/export

我可以成功地培训模型：

data = pandas.read_csv(_pathToData).dropna()
dataset = data[['description', 'class']].copy()

training_input_function = tensorflow.estimator.inputs.pandas_input_fn(
   dataset,
   dataset['class'],
   num_epochs = None,
   shuffle = True
)

_estimator.train(
   input_fn = training_input_function,
   steps = 1000
)

用它来预测事情：

descriptions = pandas.DataFrame(['sales', 'equipment'], columns=['description'])

predict_input_fn = tensorflow.estimator.inputs.pandas_input_fn(
   descriptions,
   shuffle = False
)

_estimator.predict(input_fn = predict_input_fn)

并使用

export\u saved模型导出模型：

feature_columns = _create_feature_columns()
feature_spec = tensorflow.feature_column.make_parse_example_spec(feature_columns)

serving_input_receiver_fn = tensorflow.estimator.export.build_parsing_serving_input_receiver_fn(feature_spec)

_estimator.export_savedmodel(
   export_dir_base = _export_location,
   serving_input_receiver_fn = serving_input_receiver_fn,
   strip_default_attrs = True,
   as_text = True
)

服务器使用以下命令运行：

def _initialise_estimator():

   feature_columns = _create_feature_columns()

   # Set run configs. Only keep one checkpoint.
   config = tensorflow.estimator.RunConfig(
      keep_checkpoint_max = 1,
   )

   # Create the estimator.
   estimator = tensorflow.estimator.DNNClassifier(
      hidden_units = [500, 100],
      feature_columns = feature_columns,
      n_classes = 98,
      optimizer = tensorflow.train.AdagradOptimizer(learning_rate = 0.003),
      model_dir = _model_location,
      config = config
   )

   return estimator


def _create_feature_columns():
   # Feature column. Assumed to be called description.
   return [tensorflow_hub.text_embedding_column(
      key = 'description',
      module_spec = 'https://tfhub.dev/google/nnlm-en-dim128/1'
   )]

_estimator = _initialise_estimator()

tensorflow_model_server --rest_api_port=8501 --model_name=prediction --model_base_path=/app/prediction/lib/export

它成功地启动了服务器

但是，当我尝试使用以下命令查询服务器时：

curl -d '{"instances": ["sales"]}' -X POST http://localhost:8501/v1/models/prediction:predict

我得到以下错误：

{“错误”：“无法分析示例输入，值：\'sales\'\n\t[{{node ParseExample/ParseExample}}}=ParseExample[Ndense=1，Nsparse=0，Tdense=[DT\u STRING]，[u output\u shapes=[？，1]]，[u-device=\”/job:localhost/replica:0/task:0/device:CPU:0\]（_arg_input_example_tensor_0_0，ParseExample/ParseExample/names，ParseExample/ParseExample/dense_keys_0，ParseExample/ParseExample/names）]”）

我做错了什么？有人知道实例数组应该采用什么格式吗？我一辈子都不能让它工作。我怀疑这与我如何使用
export\u savedmodel
保存模型有关，但我不确定

非常感谢您的帮助。
我也有同样的问题。。。。