def start_game(self, map_type):
self.agents = [Player(self, 'green', pygame.K_d, pygame.K_s, pygame.K_a, pygame.K_w, pygame.K_f, pygame.K_g, 100, 100, 180),
Player(self, 'purple', pygame.K_RIGHT, pygame.K_DOWN, pygame.K_LEFT, pygame.K_UP, pygame.K_k, pygame.K_l, 900, 600)]
self.map = map.World(self, map_type)
self.map.generate()
self.stats = Stats(*self.agents)
self.state = S_GAME
def continue_game(self):
self.agents = [Player(self, 'green', pygame.K_d, pygame.K_s, pygame.K_a, pygame.K_w, pygame.K_f, pygame.K_g, 100, 100),
Player(self, 'purple', pygame.K_RIGHT, pygame.K_DOWN, pygame.K_LEFT, pygame.K_UP, pygame.K_k, pygame.K_l, 900, 600)]
self.agents[0].name = self.all_player_names[0]
self.agents[1].name = self.all_player_names[1]
self.map = map.World(self, random.choice(self.maps))
self.map.generate()
self.state = S_GAME
def new_agent(self, pos, strategy):
"""
Method that creates a new agent, and adds it to the correct scheduler.
"""
agent = Player(self.next_id(), self, pos, strategy)
self.grid.place_agent(agent, pos)
self.scheduler.add(agent)
def eval(model, othermodel):
wins = 0
for evaluator in evaluators:
evaluator.player = model
evaluator.opponent = othermodel
results = Parallel(n_jobs=NUM_CORES)(
delayed(getRes)(evaluators[i], i < NUM_CORES / 2)
for i in range(NUM_CORES))
print(results)
wins = np.sum(results)
print(argv[1] + ' wins ' + str(wins) + ' games against ' + argv[2])
models = []
for a in argv[1:]:
print(a)
if a == 'manual':
continue
if a == 'random':
models.append(Player().to(DEVICE))
else:
models.append(torch.load(a).to(DEVICE))
if argv[2] == 'manual':
models.append(Player().to(DEVICE))
models[0].eval()
models[1].eval()
eval(models[0], models[1])
print(DEVICE, num_cores)
vHistory = []
pHistory = []
fig, ax1 = plt.subplots()
ax1.set_xlabel('iterations * 100')
ax1.set_ylabel('Value Loss', color='tab:red')
ax1.tick_params(axis='y', labelcolor='tab:red')
ax2 = ax1.twinx()
ax2.set_ylabel('Policy Loss', color='tab:blue')
ax2.tick_params(axis='y', labelcolor='tab:blue')
if path.exists(BEST_PATH):
alphazero = torch.load(BEST_PATH)
else:
alphazero = Player().to(DEVICE)
torch.save(alphazero, BEST_PATH)
simulators = [Game(alphazero, mctsEnable=True) for c in range(num_cores)]
def genGame(sim, movelimit):
localdf = pd.DataFrame({
"States": [],
"Actions": [],
"ActionScores": [],
"Rewards": [],
"Done": []
})
sim.player = alphazero
for i in range(int(GAMES / num_cores)):