def run_game(self, game_state: GameState, game_id, debug=False):
start_time = time.time()
while not game_state.is_game_over():
next_state, next_score = self.take_turn(game_state)
# Back propagate for new weights.
self.back_propagate(game_state.to_array(), next_score)
game_state = next_state
if debug:
print(f"moved to state with score {best_score}")
end_time = time.time()
if debug:
print(f"Game {game_id} ended with {game_state.has_player_won()}!"
f" Turns: {game_state.turns}, Time: {end_time-start_time}s")
return game_state.has_player_won(), game_state.turns
model.add(Dense(10, init='uniform', activation='relu'))
model.add(Dense(1, init='uniform', activation='linear'))
model.compile(loss='mse', optimizer='adam', metrics=['accuracy'])
# Parameters
D = deque() # Register where the actions will be stored
observetime = 500 # Number of timesteps we will be acting on the game and observing results
epsilon = 0.7 # Probability of doing a random move
gamma = 0.9 # Discounted future reward. How much we care about steps further in time
mb_size = 500 # Learning minibatch size
# FIRST STEP: Knowing what each action does (Observing)
observation = GameState(["Michael"]) # Game begins
state = observation.to_array()
done = False
turns = 0
for t in range(observetime):
possible_moves = observation.get_possible_moves()[0]
if (DEBUG):
if t < 10 or t % 20 == 0:
print("t = " + str(t))
print(observation)
print("len = " + str(len(possible_moves)))
if (t % 100 == 0 and turns >= 100):
observation = GameState(["Michael"]) # Game begins
state = observation.to_array()
done = False
turns = 0
print("RESTART: t = " + str(t))