使用aiplatform_v1异步预测时遇实例不可迭代错误求助

尝试将Python Dict传入aiplatform_v1.PredictRequest，通过aiplatform_v1.PredictionServiceAsyncClient执行预测操作时，出现错误：

TypeError: Argument for field google.cloud.aiplatform.v1.PredictRequest.instances is not iterable

已参考官方示例将Python Dict解析为google.protobuf.Value并包装成列表，但仍报错。以下是相关代码：

from google.protobuf.struct_pb2 import Value, Struct, ListValue
from google.protobuf import json_format
from google.oauth2 import service_account
from google.api_core import exceptions
from google.cloud import aiplatform_v1

async def model_predict(json_data):

    # Load credentials from a service account key file
    key_path = 'path-to-key-file.json'
    credentials = service_account.Credentials.from_service_account_file(
        key_path,
        scopes=['https://www.googleapis.com/auth/cloud-platform']
    )
   
    # Set up the endpoint configs and pass user input
    project_id = "XXXXXX"
    endpoint_id = "XXXXXXX"
    instance_dict = json_data
    location = "us-central1"
    api_endpoint = "us-central1-aiplatform.googleapis.com"
    
    # Format user input to pass into request
    instance = json_format.ParseDict(instance_dict, Value())
    parameters_dict = {}
    parameters= json_format.ParseDict(parameters_dict, Value())

    # Set up async prediction service client 
    client_options = {"api_endpoint": api_endpoint}

    prediction_client = aiplatform_v1.PredictionServiceAsyncClient(
        credentials=credentials, client_options=client_options
    )

    # Build Prediction request
    endpoint_path = prediction_client.endpoint_path(
        project=project_id, location=location, endpoint=endpoint_id,
    )

    request = aiplatform_v1.PredictRequest(
        endpoint=endpoint_path, instances= [instance], parameters = parameters
    )

    # Call the predict method asynchronously
    try:
        response = await prediction_client.predict(request=request)
        print(response)
        predictions = response.predictions

        return predictions, None

    except exceptions.GoogleAPIError as error:
        error_message = f"Prediction request failed: {error}"
        return None, error_message

打印的传入实例类型及样本：

<class 'google.protobuf.struct_pb2.Value'> struct_value { fields { key: "Ticket_Subject" value { string_value: "Battery Issue" } } fields { key: "Ticket_Status" value { string_value: "Open" } } fields { key: "Ticket_Priority" value { string_value: "High" } } .... (more fields structured as above) }

解决方法 #

方案1：直接使用Python原生类型（推荐） #

Google Cloud AI Platform的客户端库支持直接传入Python原生字典/列表，会自动完成与protobuf结构的转换，避免手动转换可能引发的类型问题。修改代码如下：

1. 替换实例和参数的构造代码：

# 直接用原始字典构造实例列表，无需手动转protobuf
instances = [json_data]
# 空参数直接传空字典
parameters = {}

2. 构建请求时传入原生类型：

request = aiplatform_v1.PredictRequest(
    endpoint=endpoint_path, instances=instances, parameters=parameters
)

方案2：手动构造正确的protobuf结构（若必须手动转换） #

如果需要手动处理protobuf结构，确保实例列表是正确的可迭代protobuf对象：

from google.protobuf.struct_pb2 import Struct, ListValue

# 构造包含实例的ListValue
instances_list = ListValue()
# 将字典解析为Struct并添加到列表中
instance_struct = json_format.ParseDict(json_data, Struct())
instances_list.values.add(struct_value=instance_struct)

# 构建请求时传入ListValue
request = aiplatform_v1.PredictRequest(
    endpoint=endpoint_path, instances=instances_list, parameters=parameters
)

验证 #

两种方案都能解决instances不可迭代的错误，其中方案1更简洁，且减少了手动转换的潜在问题。

内容的提问来源于stack exchange，提问作者jaspercoleman