AWS DeepAR预测返回400错误问题排查求助

尝试使用AWS DeepAR基于当前及过往月份的数据，对时间序列的下15条记录进行预测。已参照官方示例完成配置，但调用predict方法时返回400 ModelError，报错信息为"Unable to evaluate payload provided"。尝试移除serializer/deserializer后，仍出现格式相关错误。

部署代码 #

predictor = estimator.deploy(
    initial_instance_count=1,
    instance_type='ml.m5.large',   
    serializer=JSONSerializer(),
    deserializer=JSONDeserializer()
    )

json_request = json.dumps({
    "instances": ts,
    "configuration": {
        "num_samples": 10,
        "output_types": ["quantiles", "samples"],
        "quantiles": ['0.2', '0.5', '0.8']
    }
})

prediction = predictor.predict(json_request)

请求JSON示例 #

{"instances": 
    [{"start": "2024-03-01", 
      "target":[60,10,86,62,21,25,7,79,33,82,34,43,14,99,5,37,85,84,88,25,2,14,15,98,14,75,70,99,12]
      }, 
     {"start": "2024-04-01", 
      "target": [55,89,40,81,87,7,49,77,37,42,48,27,89,45,85]
      }], 
 "configuration": {"num_samples": 15, "output_types": ["quantiles", "samples"], "quantiles": ["0.2", "0.5", "0.8"]}}

报错堆栈 #

---------------------------------------------------------------------------
ModelError                                Traceback (most recent call last)
Cell In[22], line 2
      1 print(type('json_request'))
----> 2 prediction = predictor.predict(json_request)
      3 print(prediction)

File c:\Users\civan\PycharmProjects\JupyterBooks\.venv\Lib\site-packages\sagemaker\base_predictor.py:212, in Predictor.predict(self, data, initial_args, target_model, target_variant, inference_id, custom_attributes, component_name)
    209 if inference_component_name:
    210     request_args["InferenceComponentName"] = inference_component_name
--> 212 response = self.sagemaker_session.sagemaker_runtime_client.invoke_endpoint(**request_args)
    213 return self._handle_response(response)

File c:\Users\civan\PycharmProjects\JupyterBooks\.venv\Lib\site-packages\botocore\client.py:553, in ClientCreator._create_api_method.<locals>._api_call(self, *args, **kwargs)
    549     raise TypeError(
    550         f"{py_operation_name}() only accepts keyword arguments."
    551     )
    552 # The "self" in this scope is referring to the BaseClient.
--> 553 return self._make_api_call(operation_name, kwargs)

File c:\Users\civan\PycharmProjects\JupyterBooks\.venv\Lib\site-packages\botocore\client.py:1009, in BaseClient._make_api_call(self, operation_name, api_params)
   1005     error_code = error_info.get("QueryErrorCode") or error_info.get(
   1006         "Code"
   1007     )
   1008     error_class = self.exceptions.from_code(error_code)
--> 1009     raise error_class(parsed_response, operation_name)
   1010 else:
   1011     return parsed_response

ModelError: An error occurred (ModelError) when calling the InvokeEndpoint operation: Received client error (400) from primary with message "Unable to evaluate payload provided".

1. 修复双重序列化问题 #

代码中同时使用json.dumps()手动序列化，又配置了JSONSerializer()，导致请求数据被双重序列化（JSON字符串被再次转义），模型无法解析：

• 解决：直接传入Python字典给predict方法，去掉json.dumps()：

  prediction = predictor.predict({
      "instances": ts,
      "configuration": {
          "num_samples": 10,
          "output_types": ["quantiles", "samples"],
          "quantiles": ['0.2', '0.5', '0.8']
      }
  })
  

2. 匹配时间序列频率 #

DeepAR要求训练与预测的时间序列频率完全一致：

• 检查训练时指定的freq参数（如'D'代表日度），确认请求中start日期+target长度对应的周期数是否匹配。示例中3月只有29条数据，若freq为'D'则缺失2条，会触发格式错误。
• 解决：补全缺失时间点的数据（用NaN填充），确保每个序列的长度与对应时间段的频率周期数一致。

3. 校验配置参数合法性 #

检查configuration字段的参数格式：

• num_samples需为正整数，quantiles取值需在0-1之间（字符串或数字类型均可，但需统一）。
• 若训练时使用了动态特征，预测请求需补充对应的dynamic_feat字段。

4. 确保端点与模型兼容性 #

• 确认部署的模型是DeepAR estimator训练生成的，未混用其他模型类型。
• 若移除序列化器后仍报错，显式指定请求的Content-Type：

  prediction = predictor.predict(data_dict, initial_args={"ContentType": "application/json"})
  

内容的提问来源于stack exchange，提问作者Claudiu