client.py

# Client process

# Repeatedly does the following:
# Gets latest neural network from server, plays many games of self play 
# with the MCTS algorithm, thereby generating training data, which it
# reports to the server upon termination of the game.

import model
import config
import selfplay

import requests
import tempfile

net = model.Model()

def load_model_weights():
    data = requests.get(config.server_address + "/weights")
    tfile, tname = tempfile.mkstemp(".h5")
    open(tname, "wb").write(data.content)
    net.load(tname)

def play_game():
    # use MCTS to play a game, and return training data
    # encode the data in a string format
    mcts = selfplay.MCTS(net)
    game_history = mcts.selfplay()
    history_str = mcts.encode_history(game_history)
    return history_str

def upload_data(string):
    # send back training data to server
    resp = requests.post(config.server_address + "/train", data=string.encode())
    assert(resp.content == b'"OK"\n')

for i in range(1):
    load_model_weights()
    for i in range(config.client_play_games_num):
        data = play_game()
        upload_data(data)