replace two isolated player process by two different set of variables in the tf graph
This commit is contained in:
parent
f425085e0a
commit
5482815de6
@ -29,11 +29,8 @@ class Game:
|
|||||||
currently leave it untouched for interacting with Go UI.
|
currently leave it untouched for interacting with Go UI.
|
||||||
'''
|
'''
|
||||||
|
|
||||||
def __init__(self, name=None, role=None, debug=False, black_checkpoint_path=None, white_checkpoint_path=None):
|
def __init__(self, name=None, debug=False, black_checkpoint_path=None, white_checkpoint_path=None):
|
||||||
self.name = name
|
self.name = name
|
||||||
if role is None:
|
|
||||||
raise ValueError("Need a role!")
|
|
||||||
self.role = role
|
|
||||||
self.debug = debug
|
self.debug = debug
|
||||||
if self.name == "go":
|
if self.name == "go":
|
||||||
self.size = 9
|
self.size = 9
|
||||||
@ -41,7 +38,7 @@ class Game:
|
|||||||
self.history_length = 8
|
self.history_length = 8
|
||||||
self.history = []
|
self.history = []
|
||||||
self.history_set = set()
|
self.history_set = set()
|
||||||
self.game_engine = go.Go(size=self.size, komi=self.komi, role=self.role)
|
self.game_engine = go.Go(size=self.size, komi=self.komi)
|
||||||
self.board = [utils.EMPTY] * (self.size ** 2)
|
self.board = [utils.EMPTY] * (self.size ** 2)
|
||||||
elif self.name == "reversi":
|
elif self.name == "reversi":
|
||||||
self.size = 8
|
self.size = 8
|
||||||
@ -76,20 +73,22 @@ class Game:
|
|||||||
self.komi = k
|
self.komi = k
|
||||||
|
|
||||||
def think(self, latest_boards, color):
|
def think(self, latest_boards, color):
|
||||||
if color == +1:
|
if color == utils.BLACK:
|
||||||
role = 'black'
|
role = 'black'
|
||||||
if color == -1:
|
elif color == utils.WHITE:
|
||||||
role = 'white'
|
role = 'white'
|
||||||
|
else:
|
||||||
|
raise ValueError("game.py[think] - unknown color : {}".format(color))
|
||||||
evaluator = lambda state:self.model(role, state)
|
evaluator = lambda state:self.model(role, state)
|
||||||
mcts = MCTS(self.game_engine, evaluator, [latest_boards, color],
|
mcts = MCTS(self.game_engine, evaluator, [latest_boards, color],
|
||||||
self.size ** 2 + 1, role=self.role, debug=self.debug, inverse=True)
|
self.size ** 2 + 1, role=role, debug=self.debug, inverse=True)
|
||||||
mcts.search(max_step=100)
|
mcts.search(max_step=100)
|
||||||
if self.debug:
|
if self.debug:
|
||||||
file = open("mcts_debug.log", 'ab')
|
file = open("mcts_debug.log", 'ab')
|
||||||
np.savetxt(file, mcts.root.Q, header="\n" + self.role + " Q value : ", fmt='%.4f', newline=", ")
|
np.savetxt(file, mcts.root.Q, header="\n" + role + " Q value : ", fmt='%.4f', newline=", ")
|
||||||
np.savetxt(file, mcts.root.W, header="\n" + self.role + " W value : ", fmt='%.4f', newline=", ")
|
np.savetxt(file, mcts.root.W, header="\n" + role + " W value : ", fmt='%.4f', newline=", ")
|
||||||
np.savetxt(file, mcts.root.N, header="\n" + self.role + " N value : ", fmt="%d", newline=", ")
|
np.savetxt(file, mcts.root.N, header="\n" + role + " N value : ", fmt="%d", newline=", ")
|
||||||
np.savetxt(file, mcts.root.prior, header="\n" + self.role + " prior : ", fmt='%.4f', newline=", ")
|
np.savetxt(file, mcts.root.prior, header="\n" + role + " prior : ", fmt='%.4f', newline=", ")
|
||||||
file.close()
|
file.close()
|
||||||
temp = 1
|
temp = 1
|
||||||
prob = mcts.root.N ** temp / np.sum(mcts.root.N ** temp)
|
prob = mcts.root.N ** temp / np.sum(mcts.root.N ** temp)
|
||||||
@ -140,6 +139,6 @@ class Game:
|
|||||||
|
|
||||||
|
|
||||||
if __name__ == "__main__":
|
if __name__ == "__main__":
|
||||||
game = Game(name="reversi", role="black", checkpoint_path=None)
|
game = Game(name="reversi", checkpoint_path=None)
|
||||||
game.debug = True
|
game.debug = True
|
||||||
game.think_play_move(utils.BLACK)
|
game.think_play_move(utils.BLACK)
|
||||||
|
@ -18,7 +18,6 @@ class Go:
|
|||||||
def __init__(self, **kwargs):
|
def __init__(self, **kwargs):
|
||||||
self.size = kwargs['size']
|
self.size = kwargs['size']
|
||||||
self.komi = kwargs['komi']
|
self.komi = kwargs['komi']
|
||||||
self.role = kwargs['role']
|
|
||||||
|
|
||||||
def _flatten(self, vertex):
|
def _flatten(self, vertex):
|
||||||
x, y = vertex
|
x, y = vertex
|
||||||
@ -332,7 +331,7 @@ class Go:
|
|||||||
|
|
||||||
|
|
||||||
if __name__ == "__main__":
|
if __name__ == "__main__":
|
||||||
go = Go(size=9, komi=3.75, role = utils.BLACK)
|
go = Go(size=9, komi=3.75)
|
||||||
endgame = [
|
endgame = [
|
||||||
1, 0, 1, 0, 1, 1, -1, 0, -1,
|
1, 0, 1, 0, 1, 1, -1, 0, -1,
|
||||||
1, 1, 1, 1, 1, 1, -1, -1, -1,
|
1, 1, 1, 1, 1, 1, -1, -1, -1,
|
||||||
|
@ -1,10 +1,10 @@
|
|||||||
import argparse
|
import argparse
|
||||||
import subprocess
|
|
||||||
import sys
|
import sys
|
||||||
import re
|
import re
|
||||||
import Pyro4
|
|
||||||
import time
|
import time
|
||||||
import os
|
import os
|
||||||
|
from game import Game
|
||||||
|
from engine import GTPEngine
|
||||||
import utils
|
import utils
|
||||||
from time import gmtime, strftime
|
from time import gmtime, strftime
|
||||||
|
|
||||||
@ -48,65 +48,13 @@ if __name__ == '__main__':
|
|||||||
if args.white_weight_path is not None and (not os.path.exists(args.white_weight_path)):
|
if args.white_weight_path is not None and (not os.path.exists(args.white_weight_path)):
|
||||||
raise ValueError("Can't find the network weights for white player.")
|
raise ValueError("Can't find the network weights for white player.")
|
||||||
|
|
||||||
# kill the old server
|
game = Game(name=args.game,
|
||||||
# kill_old_server = subprocess.Popen(['killall', 'pyro4-ns'])
|
black_checkpoint_path=args.black_weight_path,
|
||||||
# print "kill the old pyro4 name server, the return code is : " + str(kill_old_server.wait())
|
white_checkpoint_path=args.white_weight_path,
|
||||||
# time.sleep(1)
|
debug=args.debug)
|
||||||
|
engine = GTPEngine(game_obj=game, name='tianshou', version=0)
|
||||||
# start a name server if no name server exists
|
|
||||||
if len(os.popen('ps aux | grep pyro4-ns | grep -v grep').readlines()) == 0:
|
|
||||||
start_new_server = subprocess.Popen(['pyro4-ns', '&'])
|
|
||||||
print("Start Name Sever : " + str(start_new_server.pid)) # + str(start_new_server.wait())
|
|
||||||
time.sleep(1)
|
|
||||||
|
|
||||||
# start two different player with different network weights.
|
|
||||||
server_list = subprocess.check_output(['pyro4-nsc', 'list'])
|
|
||||||
current_time = strftime("%Y%m%d_%H%M%S", gmtime())
|
|
||||||
|
|
||||||
black_role_name = 'black' + current_time
|
|
||||||
white_role_name = 'white' + current_time
|
|
||||||
|
|
||||||
black_player = subprocess.Popen(
|
|
||||||
['python', '-u', 'player.py', '--game=' + args.game, '--role=' + black_role_name,
|
|
||||||
'--checkpoint_path=' + str(args.black_weight_path), '--debug=' + str(args.debug)],
|
|
||||||
stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
|
|
||||||
bp_output = black_player.stdout.readline()
|
|
||||||
bp_message = bp_output
|
|
||||||
# '' means player.py failed to start, "Start requestLoop" means player.py start successfully
|
|
||||||
while bp_output != '' and "Start requestLoop" not in bp_output:
|
|
||||||
bp_output = black_player.stdout.readline()
|
|
||||||
bp_message += bp_output
|
|
||||||
print("============ " + black_role_name + " message ============" + "\n" + bp_message),
|
|
||||||
|
|
||||||
white_player = subprocess.Popen(
|
|
||||||
['python', '-u', 'player.py', '--game=' + args.game, '--role=' + white_role_name,
|
|
||||||
'--checkpoint_path=' + str(args.white_weight_path), '--debug=' + str(args.debug)],
|
|
||||||
stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
|
|
||||||
wp_output = white_player.stdout.readline()
|
|
||||||
wp_message = wp_output
|
|
||||||
while wp_output != '' and "Start requestLoop" not in wp_output:
|
|
||||||
wp_output = white_player.stdout.readline()
|
|
||||||
wp_message += wp_output
|
|
||||||
print("============ " + white_role_name + " message ============" + "\n" + wp_message),
|
|
||||||
|
|
||||||
server_list = ""
|
|
||||||
while (black_role_name not in server_list) or (white_role_name not in server_list):
|
|
||||||
if python_version < (3, 0):
|
|
||||||
# TODO : @renyong what is the difference between those two options?
|
|
||||||
server_list = subprocess.check_output(['pyro4-nsc', 'list'])
|
|
||||||
else:
|
|
||||||
server_list = subprocess.check_output(['pyro4-nsc', 'list'])
|
|
||||||
print("Waiting for the server start...")
|
|
||||||
time.sleep(1)
|
|
||||||
print(server_list)
|
|
||||||
print("Start black player at : " + str(black_player.pid))
|
|
||||||
print("Start white player at : " + str(white_player.pid))
|
|
||||||
|
|
||||||
data = Data()
|
data = Data()
|
||||||
player = [None] * 2
|
|
||||||
player[0] = Pyro4.Proxy("PYRONAME:" + black_role_name)
|
|
||||||
player[1] = Pyro4.Proxy("PYRONAME:" + white_role_name)
|
|
||||||
|
|
||||||
role = ["BLACK", "WHITE"]
|
role = ["BLACK", "WHITE"]
|
||||||
color = ['b', 'w']
|
color = ['b', 'w']
|
||||||
|
|
||||||
@ -119,7 +67,7 @@ if __name__ == '__main__':
|
|||||||
game_num = 0
|
game_num = 0
|
||||||
try:
|
try:
|
||||||
while True:
|
while True:
|
||||||
# while game_num < evaluate_rounds:
|
#while game_num < evaluate_rounds:
|
||||||
start_time = time.time()
|
start_time = time.time()
|
||||||
num = 0
|
num = 0
|
||||||
pass_flag = [False, False]
|
pass_flag = [False, False]
|
||||||
@ -127,7 +75,7 @@ if __name__ == '__main__':
|
|||||||
# end the game if both palyer chose to pass, or play too much turns
|
# end the game if both palyer chose to pass, or play too much turns
|
||||||
while not (pass_flag[0] and pass_flag[1]) and num < size[args.game] ** 2 * 2:
|
while not (pass_flag[0] and pass_flag[1]) and num < size[args.game] ** 2 * 2:
|
||||||
turn = num % 2
|
turn = num % 2
|
||||||
board = player[turn].run_cmd(str(num) + ' show_board')
|
board = engine.run_cmd(str(num) + ' show_board')
|
||||||
board = eval(board[board.index('['):board.index(']') + 1])
|
board = eval(board[board.index('['):board.index(']') + 1])
|
||||||
for i in range(size[args.game]):
|
for i in range(size[args.game]):
|
||||||
for j in range(size[args.game]):
|
for j in range(size[args.game]):
|
||||||
@ -135,7 +83,7 @@ if __name__ == '__main__':
|
|||||||
print "\n",
|
print "\n",
|
||||||
data.boards.append(board)
|
data.boards.append(board)
|
||||||
start_time = time.time()
|
start_time = time.time()
|
||||||
move = player[turn].run_cmd(str(num) + ' genmove ' + color[turn])[:-1]
|
move = engine.run_cmd(str(num) + ' genmove ' + color[turn])[:-1]
|
||||||
print("\n" + role[turn] + " : " + str(move)),
|
print("\n" + role[turn] + " : " + str(move)),
|
||||||
num += 1
|
num += 1
|
||||||
match = re.search(pattern, move)
|
match = re.search(pattern, move)
|
||||||
@ -147,21 +95,19 @@ if __name__ == '__main__':
|
|||||||
# print "no match"
|
# print "no match"
|
||||||
play_or_pass = ' PASS'
|
play_or_pass = ' PASS'
|
||||||
pass_flag[turn] = True
|
pass_flag[turn] = True
|
||||||
result = player[1 - turn].run_cmd(str(num) + ' play ' + color[turn] + ' ' + play_or_pass + '\n')
|
prob = engine.run_cmd(str(num) + ' get_prob')
|
||||||
prob = player[turn].run_cmd(str(num) + ' get_prob')
|
|
||||||
prob = space.sub(',', prob[prob.index('['):prob.index(']') + 1])
|
prob = space.sub(',', prob[prob.index('['):prob.index(']') + 1])
|
||||||
prob = prob.replace('[,', '[')
|
prob = prob.replace('[,', '[')
|
||||||
prob = prob.replace('],', ']')
|
prob = prob.replace('],', ']')
|
||||||
prob = eval(prob)
|
prob = eval(prob)
|
||||||
data.probs.append(prob)
|
data.probs.append(prob)
|
||||||
score = player[0].run_cmd(str(num) + ' get_score')
|
score = engine.run_cmd(str(num) + ' get_score')
|
||||||
print("Finished : {}".format(score.split(" ")[1]))
|
print("Finished : {}".format(score.split(" ")[1]))
|
||||||
if eval(score.split(" ")[1]) > 0:
|
if eval(score.split(" ")[1]) > 0:
|
||||||
data.winner = utils.BLACK
|
data.winner = utils.BLACK
|
||||||
if eval(score.split(" ")[1]) < 0:
|
if eval(score.split(" ")[1]) < 0:
|
||||||
data.winner = utils.WHITE
|
data.winner = utils.WHITE
|
||||||
player[0].run_cmd(str(num) + ' clear_board')
|
engine.run_cmd(str(num) + ' clear_board')
|
||||||
player[1].run_cmd(str(num) + ' clear_board')
|
|
||||||
file_list = os.listdir(args.data_path)
|
file_list = os.listdir(args.data_path)
|
||||||
current_time = strftime("%Y%m%d_%H%M%S", gmtime())
|
current_time = strftime("%Y%m%d_%H%M%S", gmtime())
|
||||||
if os.path.exists(args.data_path + current_time + ".pkl"):
|
if os.path.exists(args.data_path + current_time + ".pkl"):
|
||||||
@ -173,7 +119,3 @@ if __name__ == '__main__':
|
|||||||
game_num += 1
|
game_num += 1
|
||||||
except KeyboardInterrupt:
|
except KeyboardInterrupt:
|
||||||
pass
|
pass
|
||||||
|
|
||||||
subprocess.call(["kill", "-9", str(black_player.pid)])
|
|
||||||
subprocess.call(["kill", "-9", str(white_player.pid)])
|
|
||||||
print("Kill all player, finish all game.")
|
|
||||||
|
@ -1,42 +0,0 @@
|
|||||||
import argparse
|
|
||||||
import Pyro4
|
|
||||||
from game import Game
|
|
||||||
from engine import GTPEngine
|
|
||||||
|
|
||||||
@Pyro4.expose
|
|
||||||
class Player(object):
|
|
||||||
"""
|
|
||||||
This is the class which defines the object called by Pyro4 (Python remote object).
|
|
||||||
It passes the command to our engine, and return the result.
|
|
||||||
"""
|
|
||||||
def __init__(self, **kwargs):
|
|
||||||
self.role = kwargs['role']
|
|
||||||
self.engine = kwargs['engine']
|
|
||||||
|
|
||||||
def run_cmd(self, command):
|
|
||||||
return self.engine.run_cmd(command)
|
|
||||||
|
|
||||||
if __name__ == '__main__':
|
|
||||||
parser = argparse.ArgumentParser()
|
|
||||||
parser.add_argument("--checkpoint_path", type=str, default="None")
|
|
||||||
parser.add_argument("--role", type=str, default="unknown")
|
|
||||||
parser.add_argument("--debug", type=str, default="False")
|
|
||||||
parser.add_argument("--game", type=str, default="go")
|
|
||||||
args = parser.parse_args()
|
|
||||||
if args.checkpoint_path == 'None':
|
|
||||||
args.checkpoint_path = None
|
|
||||||
game = Game(name=args.game, role=args.role,
|
|
||||||
checkpoint_path=args.checkpoint_path,
|
|
||||||
debug=eval(args.debug))
|
|
||||||
engine = GTPEngine(game_obj=game, name='tianshou', version=0)
|
|
||||||
|
|
||||||
daemon = Pyro4.Daemon() # make a Pyro daemon
|
|
||||||
ns = Pyro4.locateNS() # find the name server
|
|
||||||
player = Player(role=args.role, engine=engine)
|
|
||||||
print("Init " + args.role + " player finished")
|
|
||||||
uri = daemon.register(player) # register the greeting maker as a Pyro object
|
|
||||||
print("Start on name " + args.role)
|
|
||||||
ns.register(args.role, uri) # register the object with a name in the name server
|
|
||||||
print("Start requestLoop " + str(uri))
|
|
||||||
daemon.requestLoop() # start the event loop of the server to wait for calls
|
|
||||||
|
|
Loading…
x
Reference in New Issue
Block a user