def reset(self, i):
self.seed = self.range_start + i % (self.range_end - self.range_start)
self.count = (self.count + 1) % (self.range_end - self.range_start)
self.packages, self.resources = gen_data.wrapper(self.seed)
self.mask = torch.zeros(1, define.get_value('package_num') + 1)
self.mask[0, -1] =1
self.path = path_obj.Path(self.resources[0], self.packages, None, False, self.device)
self.reward = 0
self.times = 0
def evaluate(model, seed):
returns = 0
packages, resources = gen_data.wrapper(seed)
path = path_obj.Path(resources[0], packages, lambda x:0, True, device)
while True:
state = path.to_state()
prob, value = model(*state)
action = torch.argmax(prob, dim=1)
action = action.cpu().numpy()[0]
package = packages[action]
times = path.getResourceNeedTime(package) + path.getResourceWorkingTime()
if times + path.getReturnTime(package) > define.get_value('time_limit'):
return returns
reward = package.getUrgency()
path.addWorkPackage(package)
path.setResourceWorkingTime(times)
path.setResourcePosition(package.getX(), package.getY(), package.getId())
returns += reward
return returns
def reset(self):
self.packages, self.resources = gen_data.wrapper(self.range_start + self.count)
self.count = (self.count + 1) % (self.range_end - self.range_start)
self.path = path_obj.Path(self.resources[0], self.packages, lambda x:0, True, self.device)
self.done = 0
return self.path.to_state()
# model.load_state_dict()
# model = model.to(device)
# model.eval()
# torch.no_grad()
define.init()
define.set_value('package_num', args.package_num)
define.set_value('time_limit', args.time_limit)
define.set_value('time_interval', args.time_interval)
define.set_value('func_type', args.func_type)
gen_data.generate_data(10000, args.package_num, args.func_type)
if args.mode == 'no':
pool = Pool(args.pool_num)
result = pool.map(f,
[gen_data.wrapper(i) for i in range(0, args.span)])
print(result)
p = [r[0][0] for r in result]
s = [r[0][1] for r in result]
print(np.mean(p, axis=0))
np.save('output/output_uniform_{}_{}'.format(args.seed, args.span), p)
pickle.dump([r[0] for r in result], open('solution/no_dqn.pkl', 'wb'))
elif args.mode == 'dqn':
state_dict = torch.load('model/model_{}.ckpt'.format(args.model),
map_location=torch.device('cpu'))
datas = [gen_data.wrapper(i) for i in range(0, args.span)]
data2s = [gen_data.get_data(i) for i in range(0, args.span)]
print(len(datas), args.span)
if args.debug:
result = []
for line in [[