update save and load

rl/simulate.py

@@ -0,0 +1,58 @@
+from tugo import rl
+from tugo.game_logic import scoring
+from tugo.game_logic import goboard_fast as goboard
+from tugo.game_logic.gotypes import Player
+
+from collections import namedtuple
+
+
+class GameRecord(namedtuple('GameRecord', 'moves winner margin')):
+    pass
+
+
+def simulate_game(black_player, white_player):
+    moves = []
+    game = goboard.GameState.new_game(19)
+    agents = {
+        Player.black: black_player,
+        Player.white: white_player,
+    }
+    while not game.is_over():
+        next_move = agents[game.next_player].select_move(game)
+        moves.append(next_move)
+        game = game.apply_move(next_move)
+
+    game_result = scoring.compute_game_result(game)
+    # print(game_result)
+
+    return GameRecord(
+        moves=moves,
+        winner=game_result.winner,
+        margin=game_result.winning_margin,
+    )
+
+
+def experience_simulation(num_games, agent1, agent2):
+    collectors = [rl.ExperienceCollector() for _ in range(2)]
+    agents = [agent1, agent2]
+
+    color1 = Player.black
+    for i in range(num_games):
+        for collector, agent in zip(collectors, agents):
+            collector.begin_episode()
+            agent.set_collector(collector)
+
+        if color1 == Player.black:
+            black_player, white_player = agent1, agent2
+        else:
+            white_player, black_player = agent2, agent1
+        game_record = simulate_game(black_player, white_player)
+
+        winner_reward = 1 if game_record.winner == color1 else -1
+        for collector in collectors:
+            collector.complete_episode(reward=winner_reward)
+            winner_reward *= -1
+
+        color1 = color1.other
+
+    return rl.combine_experience(collectors)