Microservice that does word and emoji prediction
Using Anaconda or Miniconda, create an environment with the packages listed in environment.yml with
conda env create -n nlp-service -f environment.yml
This creates a new conda environment named nlp-service. For emoji prediction support, we are using TorchMoji. Clone the TorchMoji repository into src, and follow their installation instructions for installing PyTorch and TorchMoji dependencies.
Activate the conda environment.
Server side: run python main.py to start the server.
Client side: connect to the server with SocketIO client and emit "autocomplete" with a word or phrase as an argument to get word predictions. Example for autocompleting the word "a":
from socketIO_client import SocketIO
# The code below is to simulate requests from the client
def callback_func(*args):
print("predictions: ", args)
# server running on localhost:8001
socket_client = SocketIO("localhost", 8001)
socket_client.connect()
socket_client.emit("autocomplete", "a", callback_func)
socket_client.wait_for_callbacks(seconds=1)
Will print:
predictions: ('and', 'at', 'as')
This example was taken from complete.py.
We currently have a server running on GCP with IP 34.73.165.89 on port 8001. To get a prediction or autocomplete a word, run the code above but with
socket_client = SocketIO("34.73.165.89", 8001)