small board with timer

config.py

@@ -11,7 +11,7 @@ class Config(object):
     # maintains past 2.5 games worth of turns if num_episodes=500 and rings=19
     # maintains past 1 game worth of turns if num_episodes=500 and rings=37
     # buffer_size = 100000
-    buffer_size = 100000
+    buffer_size = 150000
     epochs = 50
     lr = 0.01
 
@@ -30,8 +30,8 @@ class Config(object):
     # MCTS settings
     # 100 sims and 1000 episodes ~ 37 seconds per episode / 13 days for 30 iters
     # 25 sims and 500 episodes ~ 9 seconds per episode / 1.5 days for 30 iters
-    num_episodes = 500
-    num_sims = 25
+    num_episodes = 750
+    num_sims = 40
     c_puct = 1
     # Should be set based on game length to encourage exploration in early moves
     temp_threshold = 6

main.py

@@ -10,7 +10,7 @@
     rings = 19
     marbles = {'w': 10, 'g': 10, 'b': 10}
     win_con = [{'w': 2}, {'g': 2}, {'b': 2}, {'w': 1, 'g': 1, 'b': 1}]
-    t = 3
+    t = 5
 
     # Setup
     game = Game(rings, marbles, win_con, t)

retrain.py

@@ -7,6 +7,7 @@
 
 from mcts import MCTS
 from selfplay import SelfPlay, Arena
+import time
 
 class Coach(object):
     def __init__(self, game, model, config):
@@ -82,8 +83,14 @@ def learn(self):
             # Step 1. Generate training examples by self play with current model
             self_play = SelfPlay(self.game, self.model)
             new_examples = []
-            for _ in range(self.config.num_episodes):
+            for i in range(self.config.num_episodes):
+                start = time.time()
                 new_examples += self_play.generate_play_data()
+                now = time.time() - start
+
+                if i % 100 == 0:
+                    print 'Time to generate an episode = %i s' %now
+
             random.shuffle(new_examples)
             self.example_buffer.extend(new_examples)
             training_examples = self.examples_to_array(self.example_buffer)