update the policy

AlphaGo/.gitignore

@@ -1,2 +1,3 @@
 data
 checkpoints
+checkpoints_origin

AlphaGo/engine.py

@@ -186,8 +186,7 @@ class GTPEngine():
         return self._game.executor.get_score(), None
 
     def cmd_show_board(self, args, **kwargs):
-        self._game.show_board()
+        return self._game.board, True
-        return None, None
 
 
 if __name__ == "main":

AlphaGo/play.py

@@ -1,89 +1,70 @@
 import subprocess
 import sys
 import re
+import Pyro4
 import time
+
+#start a name server to find the remote object
+kill_old_server = subprocess.Popen(['killall', 'pyro4-ns'])
+print "kill old server, the return code is : " + str(kill_old_server.wait())
+time.sleep(1)
+start_new_server = subprocess.Popen(['pyro4-ns', '&'])
+print "Start Name Sever : " + str(start_new_server.pid)# + str(start_new_server.wait())
+time.sleep(1)
+agent_v0 = subprocess.Popen(['python', '-u', 'player.py', '--role=black'],
+                            stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+time.sleep(3)
+print "Start Player 0 at : " + str(agent_v0.pid)
+agent_v1 = subprocess.Popen(['python', '-u', 'player.py', '--role=white', '--checkpoint_path=./checkpoints_origin/'],
+                            stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+time.sleep(3)
+print "Start Player 1 at : " + str(agent_v1.pid)
+
+player = [None] * 2
+player[0] = Pyro4.Proxy("PYRONAME:black")
+player[1] = Pyro4.Proxy("PYRONAME:white")
+
+role = ["BLACK", "WHITE"]
+color = ['b', 'w']
+
 pattern = "[A-Z]{1}[0-9]{1}"
 size = 9
-agent_v1 = subprocess.Popen(['python', '-u', 'test.py'], stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
+show = ['.', 'X', 'O']
-agent_v0 = subprocess.Popen(['python', '-u', 'test.py', '--checkpoint_path=./checkpoints_origin/'], stdin=subprocess.PIPE,
-                             stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
 
-
-num = 0
 game_num = 0
-black_pass = False
+while game_num < 1:
-white_pass = False
+    num = 0
-
+    pass_flag = [False, False]
-
-while game_num < 10:
     print("Start game {}".format(game_num))
-    while not (black_pass and white_pass) and num < size ** 2 * 2:
+    # end the game if both palyer chose to pass, or play too much turns
-        print(num)
+    while not (pass_flag[0] and pass_flag[1]) and num < size ** 2 * 2:
-        if num % 2 == 0:
+        turn = num % 2
-            print('BLACK TURN')
+        move = player[turn].run_cmd(str(num) + ' genmove ' + color[turn] + '\n')
-            agent_v1.stdin.write(str(num) + ' genmove b\n')
+        print role[turn] + " : " + str(move),
-            agent_v1.stdin.flush()
+        num += 1
-            result = agent_v1.stdout.readline()
+        match = re.search(pattern, move)
-            sys.stdout.write(result)
+        if match is not None:
-            sys.stdout.flush()
+            #print "match : " + str(match.group())
-            num += 1
+            play_or_pass = match.group()
-            match = re.search(pattern, result)
+            pass_flag[turn] = False
-            print("COPY BLACK")
-            if match is not None:
-                agent_v0.stdin.write(str(num) + ' play b ' + match.group() + '\n')
-                agent_v0.stdin.flush()
-                result = agent_v0.stdout.readline()
-                sys.stdout.flush()
-            else:
-                agent_v0.stdin.write(str(num) + ' play b PASS\n')
-                agent_v0.stdin.flush()
-                result = agent_v0.stdout.readline()
-                sys.stdout.flush()
-            if re.search("pass", result) is not None:
-                black_pass = True
-            else:
-                black_pass = False
         else:
-            print('WHITE TURN')
+            #print "no match"
-            agent_v0.stdin.write(str(num) + ' genmove w\n')
+            play_or_pass = ' PASS'
-            agent_v0.stdin.flush()
+            pass_flag[turn] = True
-            result = agent_v0.stdout.readline()
+        result = player[1 - turn].run_cmd(str(num) + ' play ' + color[turn] + ' ' + play_or_pass + '\n')
-            sys.stdout.write(result)
+        board = player[turn].run_cmd(str(num) + ' show_board')
-            sys.stdout.flush()
+        board = eval(board[board.index('['):board.index(']') + 1])
-            num += 1
+        for i in range(size):
-            match = re.search(pattern, result)
+            for j in range(size):
-            print("COPY WHITE")
+                print show[board[i * size + j]] + " ",
-            if match is not None:
+            print "\n",
-                agent_v1.stdin.write(str(num) + ' play w ' + match.group() + '\n')
-                agent_v1.stdin.flush()
-                result = agent_v1.stdout.readline()
-                sys.stdout.flush()
-            else:
-                agent_v1.stdin.write(str(num) + ' play w PASS\n')
-                agent_v1.stdin.flush()
-                result = agent_v1.stdout.readline()
-                sys.stdout.flush()
-            if re.search("pass", result) is not None:
-                black_pass = True
-            else:
-                black_pass = False
 
-    print("Finished")
+    score = player[turn].run_cmd(str(num) + ' get_score')
-    print("\n")
+    print "Finished : ", score.split(" ")[1]
-
+    player[0].run_cmd(str(num) + ' clear_board')
-    agent_v1.stdin.write('clear_board\n')
+    player[1].run_cmd(str(num) + ' clear_board')
-    agent_v1.stdin.flush()
-    result = agent_v1.stdout.readline()
-    sys.stdout.flush()
-
-    agent_v0.stdin.write('clear_board\n')
-    agent_v0.stdin.flush()
-    result = agent_v0.stdout.readline()
-    sys.stdout.flush()
-
-    agent_v1.stdin.write('get_score\n')
-    agent_v1.stdin.flush()
-    result = agent_v1.stdout.readline()
-    sys.stdout.write(result)
-    sys.stdout.flush()
     game_num += 1
+
+subprocess.call(["kill", "-9", str(agent_v0.pid)])
+subprocess.call(["kill", "-9", str(agent_v1.pid)])
+print "Kill all player, finish all game."

AlphaGo/player.py

@@ -0,0 +1,37 @@
+import argparse
+import time
+import sys
+import Pyro4
+
+from game import Game
+from engine import GTPEngine
+
+@Pyro4.expose
+class Player(object):
+    def __init__(self, **kwargs):
+        self.role = kwargs['role']
+        self.engine = kwargs['engine']
+
+    def run_cmd(self, command):
+        #return "inside the Player of player.py"
+        return self.engine.run_cmd(command)
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--checkpoint_path", type=str, default="./checkpoints/")
+    parser.add_argument("--role", type=str, default="unknown")
+    args = parser.parse_args()
+
+    game = Game(checkpoint_path=args.checkpoint_path)
+    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 Request Loop " + str(uri)
+    daemon.requestLoop()                   # start the event loop of the server to wait for calls
+

AlphaGo/test.py

@@ -1,19 +0,0 @@
-import sys
-from game import Game
-from engine import GTPEngine
-# import utils
-import argparse
-import time
-
-parser = argparse.ArgumentParser()
-parser.add_argument("--checkpoint_path", type=str, default="./checkpoints/")
-args = parser.parse_args()
-
-game = Game(checkpoint_path=args.checkpoint_path)
-engine = GTPEngine(game_obj=game, name='tianshou', version=0)
-
-while not engine.disconnect:
-    command = sys.stdin.readline()
-    result = engine.run_cmd(command)
-    sys.stdout.write(result)
-    sys.stdout.flush()

README.md

@@ -46,9 +46,9 @@ Tianshou(天授) is a reinforcement learning platform. The following image illus
 
 Please follow [google python coding style](https://google.github.io/styleguide/pyguide.html)
 
-Files should all be named with lower case letters and underline.
+All files/folders should be named with lower case letters and underline (except specified names such as `AlphaGo`).
 
-Try to use full names. Don't use too many abbrevations for class/function/variable names except common abbrevations (such as `num` for number, `dim` for dimension, `env` for environment, `op` for operation). For now we use `pi` to refer to the policy in examples/ppo_example.py.
+Try to use full names. Don't use abbrevations for class/function/variable names except common abbrevations (such as `num` for number, `dim` for dimension, `env` for environment, `op` for operation). For now we use `pi` to refer to the policy in examples/ppo_example.py.
 
 The """xxx""" comment should be written right after class/function. Also comment the part that's not intuitive during the code. We must comment, but for now we don't need to polish them.
 

examples/ppo_example.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python
 
-import tensorflow as tf, numpy as np
+import tensorflow as tf
+import numpy as np
 import time
 import gym