Add --random_selfplay option to play 1 game with random agents, which

backgammon/game.py

@@ -12,6 +12,7 @@ class Game:
     OFF = 'off'
     ON = 'on'
     TOKENS = ['x', 'o']
+    SLEEP = 1
 
     def __init__(self, layout=LAYOUT, grid=None, off_pieces=None, bar_pieces=None, num_pieces=None, players=None):
         """
@@ -80,11 +81,14 @@ def next_step(self, player, player_num, draw=False):
     def take_turn(self, player, roll, draw=False):
         if draw:
             print("Player %s rolled <%d, %d>." % (player.player, roll[0], roll[1]))
-            time.sleep(1)
+            time.sleep(self.SLEEP)
 
         moves = self.get_actions(roll, player.player, nodups=True)
         move = player.get_action(moves, self) if moves else None
 
+        if draw:
+            print("Player %s played %s" % (player.player, str(move)))
+
         if move:
             self.take_action(move, player.player)
 

main.py

@@ -8,6 +8,7 @@
 
 flags.DEFINE_boolean('test', False, 'If true, test against a random strategy.')
 flags.DEFINE_boolean('play', False, 'If true, play against a trained TD-Gammon strategy.')
+flags.DEFINE_boolean('random_selfplay', False, 'Watch 2 random agents play')
 flags.DEFINE_boolean('restore', False, 'If true, restore a checkpoint before training.')
 
 model_path = os.environ.get('MODEL_PATH', 'models/')
@@ -32,5 +33,7 @@
             model.test(episodes=1000)
         elif FLAGS.play:
             model.play()
+        elif FLAGS.random_selfplay:
+            model.random_selfplay()
         else:
             model.train()

model.py

@@ -192,6 +192,12 @@ def test(self, episodes=100, draw=False):
                 winners[0], winners[1], winners_total, \
                 (winners[0] / winners_total) * 100.0))
 
+    def random_selfplay(self):
+        players = [RandomAgent(Game.TOKENS[0]), RandomAgent(Game.TOKENS[1])]
+        game = Game.new()
+        game.SLEEP = 0
+        winner = game.play(players, draw=True)
+
     def train(self):
         tf.train.write_graph(self.sess.graph_def, self.model_path, 'td_gammon.pb', as_text=False)
         summary_writer = tf.train.SummaryWriter('{0}{1}'.format(self.summary_path, int(time.time()), self.sess.graph_def))