Modified the computing of the statistics, augmented the number of simulations

- Also fixed a bug in the time computation of the MCTS algorithm

compute.py

@@ -13,11 +13,15 @@ from game.strategy import MCTSStrategy, StrategyAlphaBeta
 from game.helpers.mcts import UCB
 from game.utils import other
 
+
+# Parameters to choose from
+# The ultimate goal is to find the perfect couple
+# that will make MCTS win most of the time.
 ucb = [UCB.ucb, UCB.ucb_tuned]
-weights = np.arange(0, 10, 0.2)
+weights = np.arange(0, 20, 0.2)
 
 
-def run(ce, fucb, n=100, debug=False):
+def run(ce, fucb, n=200, debug=False):
     """
     Runs n games with the given parameters and returns the
     number of wins for each player and the total time spent
@@ -25,6 +29,7 @@ def run(ce, fucb, n=100, debug=False):
     """
     wins = [0, 0]
     ttimes = [0, 0]
+    niters = [0, 0]
 
     for _ in range(n):
         # (size, top player, bottom player)
@@ -33,7 +38,7 @@ def run(ce, fucb, n=100, debug=False):
 
         ab = StrategyAlphaBeta(grid, rules, 1)
         mcts = MCTSStrategy(
-            grid, rules, 2, simulations=10, exploration_weight=ce, ucb=fucb
+            grid, rules, 2, simulations=100, exploration_weight=ce, ucb=fucb
         )
 
         ply = 1
@@ -47,11 +52,13 @@ def run(ce, fucb, n=100, debug=False):
                 action = ab.get_action(grid, ply)
                 ttimes[ply - 1] += time.time() - s
                 grid.move(action, ply)
+                niters[ply - 1] += 1
             else:
                 s = time.time()
                 action = mcts.get_action(grid, ply)
                 ttimes[ply - 1] += time.time() - s
                 grid.move(action, ply)
+                niters[ply - 1] += 1
 
             if debug:
                 grid.debug_plot(save=True)
@@ -63,7 +70,7 @@ def run(ce, fucb, n=100, debug=False):
         else:
             wins[1] += 1
 
-    return wins, ttimes, n
+    return wins, ttimes, n, niters
 
 
 def iterate():
@@ -78,9 +85,9 @@ def iterate():
     with tqdm(total=niter, desc="Computing statistics...") as pbar:
         for i, u in enumerate(ucb):
             for j, w in enumerate(weights):
-                nwins, t, n = run(w, u)
+                nwins, t, n, niters = run(w, u)
                 wins[i, j] = nwins[1] / n * 100
-                times[i, j] = t[1] / n
+                times[i, j] = t[1] / niters[1]
 
                 pbar.update(1)
 

dogopher.py

@@ -57,6 +57,39 @@ def strategy(
     pass
 
 
+def run_one():
+    grid: HexGrid = HexGrid.split_state(4, 2, 1)
+    rules: Rules = Dodo()
+
+    ab = StrategyAlphaBeta(grid, rules, 1)
+    mcts = MCTSStrategy(
+        grid, rules, 2, simulations=100, exploration_weight=10, ucb=UCB.ucb_tuned
+    )
+
+    ply = 1
+    ttime = 0
+    niter = 0
+
+    while not rules.game_over(grid):
+        if ply == 1:
+            action = ab.get_action(grid, ply)
+            grid.move(action, ply)
+        else:
+            start = time.time()
+            action = mcts.get_action(grid, ply)
+            ttime += time.time() - start
+            grid.move(action, ply)
+            niter += 1
+
+        ply = other(ply)
+
+    return ttime / niter, rules.has_won(grid, 2)
+
+
+average, win = run_one()
+print(f"Time: {average:.4f} | Win: {win}")
+
+
 def run(n=100, debug=False):
     wins = [0, 0]
     ttimes = [0, 0]
@@ -101,7 +134,7 @@ def run(n=100, debug=False):
     return wins, ttimes, n
 
 
-w, t, n = run(100)
+# w, t, n = run(100)
 
-print(f"Alphabeta | Wins percentage: {w[0] / n * 100:.2f}% | Time: {t[0] / n:.2f}")
-print(f"MCTS      | Wins percentage: {w[1] / n * 100:.2f}% | Time: {t[1] / n:.2f}")
+# print(f"Alphabeta | Wins percentage: {w[0] / n * 100:.2f}% | Time: {t[0] / n:.2f}")
+# print(f"MCTS      | Wins percentage: {w[1] / n * 100:.2f}% | Time: {t[1] / n:.2f}")