def _runTurn(self):
for p in range(len(self.players)):
# refill player sp before running actions
self.state["player_{}_external".format(p)]["sp"] = self.state["player_{}_external".format(p)]["max_sp"]
self._runActionsForP(p)
if self.winning_player != None:
return
Stats.recordStat("turns")
self.state["g"]["turn"] += 1
def _selectAction(self, recommended_a_id=None):
# if no action is recommended or we randomly roll below our
# random_action_rate, select a random action. TODO it might be a good idea
# to have state similarity here to pick a closest observed action
possible_actions = getValidActionsInState(self.s)
possible_action_ids = [
Database.upsertAction(action) for action in possible_actions
]
if not recommended_a_id or np.random.random(
) < self.random_action_rate or recommended_a_id not in possible_action_ids:
random_action_index = np.random.randint(len(possible_action_ids))
random_action_id = possible_action_ids[random_action_index]
self._printIfVerbose("agent randomly chose",
possible_actions[random_action_index])
return random_action_id, possible_actions[random_action_index]
else:
action = Database.getAction(recommended_a_id)
self._printIfVerbose("agent chose", action)
Stats.recordStat("{}{}".format(
"chosen_action={}".format(action["action"]),
"_id={}".format(action["card_id"])
if "card_id" in action and action["card_id"] != None else ""))
return recommended_a_id, action
def main():
# TODO implement command-line args
# initialize database
Database.initialize()
q = Database.getQTable()
cards = loadCardDefinitions()
characters = loadCharacterDefinitions()
# initialize card definitions for querying
CardDefinitions.setDefinitions(cards["main"], cards["treasures"],
cards["answers"])
# how many games to play per run
num_games = run_constants["num_games"]
# every nth game will be verbose
verbose_mod = run_constants["verbose_mod"]
# game params as defined in game/game.py
game_params = {
# "num_agents"
# "num_humans"
# "max_turns"
}
# agent params as defined in player/agent.py
agent_params = {
"learning_rate": agent_constants["learning_rate"]
# "discount_factor"
# "endgame_discount_factor"
# "random_action_rate"
# "dyna_steps"
}
# deck params as defined in game/game.py
deck_params = {
"main_cards": CardDefinitions.cards["main"],
"treasure_cards": CardDefinitions.cards["treasures"],
"answer_cards": CardDefinitions.cards["answers"]
}
# character params as defined in game/game.py
character_params = {"characters": characters}
for game_number in range(num_games):
verbose = game_number % verbose_mod == verbose_mod - 1
game_params["verbose"] = verbose
agent_params["verbose"] = verbose
print("Running game {}".format(game_number + 1))
game = Game(q, game_params, agent_params, deck_params,
character_params)
game.run()
Stats.recordStat("games")
agent_params[
"learning_rate"] *= 1 - agent_constants["learning_rate_decay"]
Database.commit()
# deinitialize database
Database.destroy()
Stats.printStats()
Stats.printQStats(q)
Stats.graphChosenActionUsage()
Stats.graphTurnCountPerGame()
