识别 GCMLE 模型中所需的占位符

Question

几周前我部署了一个模型，但我忘记了该模型需要哪些输入特征（是的，我知道我应该更好地跟踪这一点）。当我运行以下命令时

gcloud ml-engine predict \
--model $MODEL_NAME \
--version v1 \
--json-instances \
../test.json

我得到以下错误：

{
  "error": "Prediction failed: Exception during model execution: AbortionError(code=StatusCode.INVALID_ARGUMENT, details=\"input size does not match signature\")"
}

我知道我的问题是因为我的 json 请求 (test.json) 中没有包含一些“必需的”占位符，但我无法找出任何方法来追溯找出缺少哪个占位符。

问题与question 相同，当发布者包含丢失的“阈值”张量时，他能够提交预测。

查看给定模型的预期输入的最简单方法是什么？

Answer 1

服务本身还不允许您查询模型的签名，因此我的建议是尽可能使用saved_model_cli（假设您没有删除原始模型）。比如：

gcloud ml-engine versions describe v1 --model census | grep deploymentUri

这将输出如下内容：

deploymentUri: gs://my_bucket/path/to/Servo/1488268526779

现在，使用saved_model_cli：

saved_model_cli show --all --dir gs://my_bucket/path/to/Servo/1488268526779

这将输出如下内容：

MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:

signature_def['serving_default']:
The given SavedModel SignatureDef contains the following input(s):
inputs['age'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1)
    name: Placeholder_8:0
inputs['capital_gain'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1)
    name: Placeholder_10:0
inputs['capital_loss'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1)
    name: Placeholder_11:0
inputs['education'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder_2:0
inputs['education_num'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1)
    name: Placeholder_9:0
inputs['gender'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder:0
inputs['hours_per_week'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1)
    name: Placeholder_12:0
inputs['marital_status'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder_3:0
inputs['native_country'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder_7:0
inputs['occupation'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder_6:0
inputs['race'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder_1:0
inputs['relationship'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder_4:0
inputs['workclass'] tensor_info:
    dtype: DT_STRING
    shape: (-1)
    name: Placeholder_5:0
The given SavedModel SignatureDef contains the following output(s):
outputs['classes'] tensor_info:
    dtype: DT_INT64
    shape: (-1)
    name: predictions/classes:0
outputs['logistic'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1, 1)
    name: predictions/logistic:0
outputs['logits'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1, 1)
    name: add:0
outputs['probabilities'] tensor_info:
    dtype: DT_FLOAT
    shape: (-1, 2)
    name: predictions/probabilities:0
Method name is: tensorflow/serving/predict

Answer 2

假设您有与您保存的模型相关联的张量流图。如果您可以打印图表（以文本格式打印 protobuffer）或使用工具进行可视化，您可以在图表中看到“signature_def”字段。这应该告诉您图表的输入和输出。您可以将您在请求中发送的张量与图表预期的输入进行比较。