This is a tensorflow serving client implementation from https://github.com/epigramai/tfserving-python-predict-client.
You can find here a Tensorflow client examples to trade.
The predict client is meant to be used with a model served by TensorFlow Serving. Because tfserving model server runs a gRPC service, it cannot be requested by just sending a normal HTTP request. The predict_client package is a grcp client that can request the service.
Read the author's blog posts about TensorFlow Serving:
and also my post serie: Deeptrading with tensorflow specially: https://todotrader.com/deep-trading-with-tensorflow-viii/
conda create --name myenv
$ conda activate myenv
(myenv)$ pip install git+https://github.com/parrondo/deeptrading-tfserving-python.git
There is one here https://github.com/bitnami/bitnami-docker-tensorflow-serving
Check out the examples.
Example 1 use fixed enter data and request a response from the serving model.
Example 2 generates random data simulating some Forex enter data and request a response from the serving model.
def init(self, host, model_name, model_version):
- host: the host (e.g. '0.0.0.0:8500')
- model_name: your model name, e.g. '07_First_Forex_Prediction'
- model_version: model version, e.g. 1.
ProdClient.predict(self, request_data, request_timeout=10):
- request_data: A list of input tensors, see the example.
- request_timeout: timeout sent to the grcp stub
from predict_client.prod_client import ProdClient
client = ProdClient('localhost:9000', 'mnist', 1)
client.predict(request_data)
The predict function returns a dictionary with keys and values for each output tensor. The values in the dictionary will have the same shapes as the output tensor's shape. If an error occurs, predict will return an empty dict.
def init(self, model_path):
- model_path
InMemoryClient.predict(self, request_data, request_timeout=None):
- request_data and request_timeout same as ProdClient, except request_timeout not used in this client.
from predict_client.inmemory_client import InMemoryClient
client = InMemoryClient('path/to/model.pb')
client.predict(request_data)
The predict function returns a dictionary with keys and values for each output tensor. The values in the dictionary will have the same shapes as the output tensor's shape. If an error occurs, predict will return an empty dict.
def init(self, mock_response):
- mock_response
MockClient.predict(self, request_data, request_timeout=None):
- request_data and request_timeout same as ProdClient, except request_timeout not used in mock client.
from predict_client.mock_client import MockClient
client = MockClient(mock_response)
client.predict(request_data)
The mock client predict function simply returns the mock response.
Caution, this is not always trivial!
pip install grpcio-tools
python -m grpc_tools.protoc -I protos/ --python_out=predict_client/pbs --grpc_python_out=predict_client/pbs protos/*