Modification and doc for unit test

tianshou/core/mcts/unit_test/README.md

@@ -0,0 +1,21 @@
+# Unit Test
+
+This is a two-player zero-sum perfect information extensive game. Player 1 and player 2 iteratively choose actions. At every iteration, player 1 players first and player 2 follows. Both players have choices 0 or 1.
+
+The number of iterations is given as a fixed number. After one game finished, the game counts the number of 0s and 1s that are choosen. If the number of 1 is more than that of 0, player 1 gets 1 and player 2 gets -1. If the number of 1 is less than that of 0, player 1 gets -1 and player 2 gets 1. Otherwise, they both get 0.
+
+## Files
+
++ game.py: run this file to play the game.
++ agent.py: a class for players. MCTS is used here.
++ ZOgame.py: the game environment.
++ mcts.py: MCTS method.
++ Evaluator: evaluator for MCTS. Rollout policy is also here.
+
+## Parameters
+
+Three paramters are given in game.py.
+
++ size: the number of iterations
++ searching_step: the number of searching times of MCTS for one step
++ temp: the temporature paramter used to tradeoff exploitation and exploration

tianshou/core/mcts/unit_test/ZOGame.py

@@ -29,7 +29,7 @@ class ZOTree:
         length = len(seq)
         if length != self.depth:
             raise ValueError("The game is not terminated!")
-	result = np.sum(seq)
+        result = np.sum(seq)
         if result > self.size:
             winner = 1
         elif result < self.size:

tianshou/core/mcts/unit_test/agent.py

@@ -4,13 +4,15 @@ import ZOGame
 import Evaluator
 from mcts import MCTS
 
-temp = 1
+
 
 
 class Agent:
-    def __init__(self, size, color):
+    def __init__(self, size, color, searching_step, temp):
         self.size = size
         self.color = color
+        self.searching_step = searching_step
+        self.temp = temp
         self.simulator = ZOGame.ZOTree(self.size)
         self.evaluator = Evaluator.rollout_policy(self.simulator, 2)
 
@@ -18,10 +20,9 @@ class Agent:
         if len(seq) >= 2 * self.size:
             raise ValueError("Game is terminated.")
         mcts = MCTS(self.simulator, self.evaluator, [seq, self.color], 2, inverse=True)
-        mcts.search(max_step=50)
+        mcts.search(max_step=self.searching_step)
         N = mcts.root.N
-        N = np.power(N, 1.0 / temp)
+        N = np.power(N, 1.0 / self.temp)
         prob = N / np.sum(N)
-        print("prob: {}".format(prob))
         action = int(np.random.binomial(1, prob[1]))
         return action

tianshou/core/mcts/unit_test/game.py

@@ -3,17 +3,19 @@ import agent
 
 
 if __name__ == '__main__':
+    size = 10
+    seaching_step = 100
+    temp = 1
     print("Our game has 2 players.")
     print("Player 1 has color 1 and plays first. Player 2 has color -1 and plays following player 1.")
     print("Both player choose 1 or 0 for an action.")
-    size = 2
     print("This game has {} iterations".format(size))
     print("If the final sequence has more 1 that 0, player 1 wins.")
     print("If the final sequence has less 1 that 0, player 2 wins.")
     print("Otherwise, both players get 0.\n")
     game = ZOGame.ZOTree(size)
-    player1 = agent.Agent(size, 1)
+    player1 = agent.Agent(size, 1, seaching_step, temp)
-    player2 = agent.Agent(size, -1)
+    player2 = agent.Agent(size, -1, seaching_step, temp)
 
     seq = []
     print("Sequence is {}\n".format(seq))

tianshou/core/mcts/unit_test/mcts.py

@@ -162,8 +162,6 @@ class MCTS(object):
             self.expansion_time += exp_time
             self.backpropagation_time += back_time
             step += 1
-            print("Q = {}".format(self.root.Q))
-            print("N = {}".format(self.root.N))
         if self.debug:
             file = open("mcts_profiling.log", "a")
             file.write("[" + str(self.role) + "]"