<type 'list'>. Valid formats: float, int, str any object that implements __iter__ or classification_pb2.ClassificationRequest

Please fill out the form below.

System Information

• Framework (e.g. TensorFlow) / Algorithm (e.g. KMeans): TensorFlow
• Framework Version: 1.11
• Python Version: 2.7
• CPU or GPU: CPU
• Python SDK Version: 1.18.3
• Are you using a custom image: No

Describe the problem

Getting the following error on inference: <type 'list'>. Valid formats: float, int, str any object that implements iter or classification_pb2.ClassificationRequest

Seems to come from here: https://github.com/aws/sagemaker-tensorflow-container/blob/ba46b9262da8b22e3242a4d35220679e6b9043c2/src/tf_container/proxy_client.py#L262

Saved Model Output

The given SavedModel SignatureDef contains the following input(s):
  inputs['inputs'] tensor_info:
      dtype: DT_STRING
      shape: (-1)
      name: input_example_tensor:0
The given SavedModel SignatureDef contains the following output(s):
  outputs['classes'] tensor_info:
      dtype: DT_STRING
      shape: (-1, 20)
      name: dnn/head/Tile:0
  outputs['scores'] tensor_info:
      dtype: DT_FLOAT
      shape: (-1, 20)
      name: dnn/head/predictions/probabilities:0
Method name is: tensorflow/serving/classify

Serving Request

model_spec {
  name: "generic_model"
  signature_name: "serve"
}
inputs {
  key: "inputs"
  value {
    dtype: DT_STRING
    tensor_shape {
      dim {
        size: 1
      }
    }
    string_val: "\n\230\361\004\n\224\361\004\n\006input\022\210\361\004\022\204\361\004\n

Code

INPUT_KEY = 'text'
OUTPUT_KEY = 'response'
ERROR_STRING = 'Requested unsupported ContentType, content type: '

def input_fn(serialized_input_data, content_type=JSON_CONTENT_TYPE):
    if content_type == JSON_CONTENT_TYPE:
        input_data = json.loads(serialized_input_data)
        input_features = transform(preprocess(input_data[INPUT_KEY]))
        return build_request("generic_model", input_features.tolist()) #input_features is a list of floats

    raise Exception(ERROR_STRING + content_type)

def build_request(name, input_ids, signature_name=DEFAULT_SERVING_SIGNATURE_DEF_KEY):
    examples = [make_example(input_ids).SerializeToString()]

    request = predict_pb2.PredictRequest()
    request.model_spec.name = name
    request.model_spec.signature_name = signature_name
    request.inputs["inputs"].CopyFrom(
        tf.contrib.util.make_tensor_proto(
            examples))
    return request

def make_example(input_ids, feature_name="input"):
    features = {
        feature_name:
            tf.train.Feature(float_list=tf.train.FloatList(value=input_ids))
    }
    return tf.train.Example(features=tf.train.Features(feature=features))

def transform(msg):
    msg = [preprocess(msg)]
    msg = TFIDF_VECTORIZER.transform(msg)
    return KBEST_SELECTOR.transform(msg.toarray()).flatten()

def preprocess(line):
    return re.sub("[^a-z0-9'.]+", " ", line.lower()).strip()

Whoops found my mistake. I adapted this code from another model I had and forgot to change the build_request function to create a classification request.

Working build_request function:

def build_request(name, input_ids, signature_name=DEFAULT_SERVING_SIGNATURE_DEF_KEY):
    examples = [make_example(input_ids)]
    request = classification_pb2.ClassificationRequest()
    request.model_spec.name = name
    request.model_spec.signature_name = signature_name
    request.input.example_list.examples.extend(examples)
    return request