Server and client for hosting ML based solutions
Installation
pip3 install myLittleServer
Server creation
Server can implement simple interface with just predict method. Use it, when you don't need to save any information which can be obtained in progress.
Example of class which fits expected interface:
class Interface:
def __init__(self, **kwargs):
pass
# data can be any object, which you expect to receive
def predict(self, data):
return dataTo run server with such interface, type this code:
from mls import Server
# Create instance of Interface class
interface = Interface()
# Server initialization
server = Server(port=3001, ml=interface)If predict is not a pure function (results of it depends on the state), than StateMachine is your way to go! Example of class which fits expected interface:
class StateMachineInterface:
# data can be any object, which you expect to receive
# method has to return events in predefined format
def update(self, data):
events = self._process(data)
return events
# data can be any object, which you expect to receive
def set_state(self, state):
self._update_my_state(state)
return NoneTo run server with such interface, type this code:
interface = StateMachineInterface()
from mls import StateMachine
# Server initialization
server = StateMachine(port=3001, ml=interface)Following syntax is the same for both Server and StateMachine.
You can start ml module initialization process inside server. It can be helpful if you want to check when server is ready. In that case pass Interface constructor and it's kwargs as separate params.
from mls import Server
# Your init kwargs
config = {'foo': 'bar'}
# Server initialization
# You need to provide class itself, not object of that class
# It's the same as if you do Interface(**ml_config)
server = Server(port=3001, ml=Interface, ml_config=config)You can turn off server logs by providing log=False:
# You can turn off logging
server = Server(port=3001, ml=interface, log=False)Client creation
Client initialization:
from mls import Client
# Client initialization
client = Client(address='http://localhost:3001')Check server status:
# Check that server launched
is_started = client.started()
# Check that ml module is set up and ready for work
is_ready = client.ready()
# You can call only ready() as it has inner check for case, when server hasn't startedResult of client call is future. So all it's methods work fine.
res = client.predict(data_to_process)Get the body of the result. It's a blocking operation:
data = res.result()Best practice is to check res for exception before running result() method as it will throw exception if one occurred on the server side.
# Otherwise res.result() will raise an exception.
if res.exception() is None:
data = res.result()You can check if call to server is finished:
while True:
if not res.done():
sleep(1)
else:
do_stuff(res.result())You can also add a callback function:
res.add_done_callback(lambda res: do_stuff(res.result()))
