Serializers#

class pai.serializers.SerializerBase#

基类：ABC

Abstract class for creating a Serializer class for predictor.

abstract serialize(data) → bytes#: Serialize the input data to bytes for transmitting.

abstract deserialize(data: bytes)#: Deserialize the data from raw bytes to Python object .

inspect_from_service(service_name: str, *, session: Optional[Session] = None)#

Inspect the online prediction service to complete the serializer instance initialization.

The implementation of the inspect_from_service method is optional. You only need to implement it if your serializer requires additional information from service metadata or if it needs to send a request to the service in order to be initialized.

class pai.serializers.JsonSerializer#

基类：SerializerBase

A Serializer object that serialize input data into JSON format and deserialize JSON formatted data into python object.

serialize(data) → bytes#: Serialize the input data to bytes for transmitting.

deserialize(data)#: Deserialize the data from raw bytes to Python object .

class pai.serializers.TensorFlowSerializer#

基类：SerializerBase

A Serializer class that responsible for transforming input/output data for TensorFlow processor service.

TensorflowSerializer initializer.

inspect_from_service(service_name: str, *, session: Optional[Session] = None)#

Inspect the service to complete serializer instance initialization.

参数:

service_name (str) -- Name of the online prediction service.
session (pai.session.Session) -- A PAI session instance used for communicating with PAI services.

classmethod inspect_model_signature_def(service_name: str, *, session: Optional[Session] = None) → Dict[str, Any]#

Inspect the TensorFlow serving model signature by sending a request to the service.

TensorFlow processor creates a prediction service and exposes an HTTP API for model signature definition.

Example API returns:

{
    "signature_name": "serving_default",
    "inputs": [
        {
            "name": "flatten_input",
            "shape": [
                -1,
                28,
                28
            ],
            "type": "DT_FLOAT"
        }
    ],
    "outputs": [
        {
            "name": "dense_1",
            "shape": [
                -1,
                10
            ],
            "type": "DT_FLOAT"
        }
    ]
}

返回:: A dictionary that represents the model signature definition.

serialize(data: Union[Dict[str, Any], PredictRequest]) → bytes#: Serialize the input data to bytes for transmitting.

deserialize(data: bytes)#: Deserialize the data from raw bytes to Python object .

class pai.serializers.PyTorchSerializer#

基类：SerializerBase

A serializer responsible for transforming input/output data for PyTorch processor service.

serialize(data: Union[ndarray, List, Tuple]) → bytes#: Serialize the input data to bytes for transmitting.

deserialize(data: bytes)#: Deserialize the data from raw bytes to Python object .

Serializers

目录

Serializers#