def feature_extractor2_numpy(sample):
resource, work_packages, total_time = sample
work_packages = [[
p.getX(),
p.getY(),
p.getUrgency(),
p.getWorkingTime(),
p.getUrgency() / p.getWorkingTime(),
p.getId()
] for p in work_packages]
gnn_feature = np.zeros((define.package_num, define.package_num, 3))
mask = np.zeros((define.package_num, define.package_num, 1))
node_feature = np.zeros((define.package_num, 3))
for p_i in work_packages:
i = p_i[5]
node_feature[i, 0] = p_i[2]
node_feature[i, 1] = p_i[3]
node_feature[i, 2] = total_time
for p_j in work_packages:
j = p_j[5]
gnn_feature[i, j, 4] = define.dis(p_i[0], p_j[0], p_i[1],
p_j[1]) / define.speed
gnn_feature[i, j, 5] = define.dis(resource[0], p_i[0], resource[1],
p_i[1]) / define.speed + p_i[3]
gnn_feature[i, j, 6] = define.dis(resource[0], p_j[0], resource[1],
p_j[1]) / define.speed + p_j[3]
mask[i, j] = 1
# for i in range(7):
# print(np.max(gnn_feature[:,:,i]))
# input()
return np.expand_dims(gnn_feature, axis=0), np.expand_dims(
mask, axis=0), np.expand_dims(node_feature, axis=0)
def feature_extractor_base(sample, device):
resource, work_packages, total_time = sample
work_packages = [[
p.getX(),
p.getY(),
p.getUrgency(),
p.getWorkingTime(),
p.getUrgency() / p.getWorkingTime(),
p.getId()
] for p in work_packages]
work_packages.sort(key=lambda x: x[5])
gnn_feature = np.zeros((define.package_num, define.package_num, 5))
resource_dis = np.zeros(define.package_num)
for p_i in work_packages:
i = p_i[5]
gnn_feature[i, :, 0] = p_i[2]
gnn_feature[:, i, 1] = p_i[2]
gnn_feature[i, :, 2] = p_i[3]
gnn_feature[:, i, 3] = p_i[3]
resource_dis[i] = define.dis(p_i[0], resource[1], p_i[1], resource[1])
for p_j in work_packages:
j = p_j[5]
gnn_feature[i, j, 4] = define.dis(p_i[0], p_j[0], p_i[1], p_j[1])
return torch.from_numpy(gnn_feature).to(device), torch.from_numpy(
resource_dis).to(device)
def feature_extractor(sample):
resource, work_packages, total_time = sample
work_packages = [[
p.getX(),
p.getY(),
p.getUrgency(),
p.getWorkingTime(),
p.getUrgency() / p.getWorkingTime()
] for p in work_packages]
work_packages.sort(key=lambda x: x[4])
work_packages.reverse()
gnn_feature = np.zeros((define.package_num, define.package_num, 8))
mask = np.zeros((define.package_num, define.package_num, 1))
for i, p_i in enumerate(work_packages):
for j, p_j in enumerate(work_packages):
gnn_feature[i, j, 0] = p_i[2]
gnn_feature[i, j, 1] = p_j[2]
gnn_feature[i, j, 2] = p_i[3]
gnn_feature[i, j, 3] = p_j[3]
gnn_feature[i, j, 4] = define.dis(p_i[0], p_j[0], p_i[1],
p_j[1]) / define.speed
gnn_feature[i, j, 5] = define.dis(resource[0], p_i[0], resource[1],
p_i[1]) / define.speed
gnn_feature[i, j, 6] = define.dis(resource[0], p_j[0], resource[1],
p_j[1]) / define.speed
gnn_feature[i, j, 7] = total_time
mask[i, j] = 1
# for i in range(7):
# print(np.max(gnn_feature[:,:,i]))
# input()
return np.expand_dims(gnn_feature, axis=0), np.expand_dims(mask, axis=0)
def extract_feat(resource, work_packages):
work_packages = [[
p.getX(),
p.getY(),
p.getUrgency(),
p.getWorkingTime(),
p.getUrgency(),
p.getId()
] for p in work_packages]
rx, ry, rid = resource.getPosition()
gnn_feature = np.zeros((define.get_value('package_num') + 1,
define.get_value('package_num') + 1, 4))
for p_i in work_packages:
i = p_i[5]
for p_j in work_packages:
j = p_j[5]
gnn_feature[i, j, 0] = p_i[2]
gnn_feature[i, j, 1] = p_j[2]
gnn_feature[
i, j,
2] = define.dis(p_i[0], p_j[0], p_i[1],
p_j[1]) / define.get_value('speed') + p_j[3]
gnn_feature[i, j, 3] = 1
for p in work_packages:
i = p[5]
gnn_feature[-1, i, 2] = define.dis(
p[0], rx, p[1], ry) / define.get_value('speed') + p[3]
gnn_feature[i, -1, 2] = define.dis(p[0], rx, p[1],
ry) / define.get_value('speed')
return torch.from_numpy(gnn_feature).unsqueeze(0).float()
def feature_extractor2(sample):
resource, work_packages, total_time, end_axis = sample
work_packages = [[
p.getX(),
p.getY(),
p.getUrgency(),
p.getWorkingTime(),
p.getUrgency(),
p.getId()
] for p in work_packages]
gnn_feature = np.zeros(
(define.get_value('package_num'), define.get_value('package_num'), 8))
mask = np.zeros(
(define.get_value('package_num'), define.get_value('package_num'), 1))
for p_i in work_packages:
i = p_i[5]
for p_j in work_packages:
j = p_j[5]
gnn_feature[i, j, 0] = p_i[2]
gnn_feature[i, j, 1] = p_j[2]
# gnn_feature[i, j, 2] = p_i[3]
# gnn_feature[i, j, 3] = p_j[3]
gnn_feature[i, j,
2] = define.dis(end_axis[0], p_i[0], end_axis[1],
p_i[1]) / define.get_value('speed')
gnn_feature[i, j,
3] = define.dis(end_axis[0], p_j[0], end_axis[1],
p_j[1]) / define.get_value('speed')
gnn_feature[
i, j,
4] = define.dis(p_i[0], p_j[0], p_i[1],
p_j[1]) / define.get_value('speed') + p_j[3]
gnn_feature[
i, j,
5] = define.dis(resource[0], p_i[0], resource[1],
p_i[1]) / define.get_value('speed') + p_i[3]
gnn_feature[
i, j,
6] = define.dis(resource[0], p_j[0], resource[1],
p_j[1]) / define.get_value('speed') + p_j[3]
gnn_feature[i, j, 7] = total_time
mask[i, j] = 1
# for i in range(7):
# print(np.max(gnn_feature[:,:,i]))
# input()
return np.expand_dims(gnn_feature, axis=0), np.expand_dims(mask, axis=0)
def feature_func(args):
gnn_feature, mask, i, p_i, total_time, work_packages, resource = args
for j, p_j in enumerate(work_packages):
gnn_feature[i, j, 0] = p_i[2]
gnn_feature[i, j, 1] = p_j[2]
gnn_feature[i, j, 2] = p_i[3] / max(total_time, 1e-3)
gnn_feature[i, j, 3] = p_j[3] / max(total_time, 1e-3)
gnn_feature[i, j, 4] = define.dis(p_i[0], p_j[0], p_i[1],
p_j[1]) / define.speed / max(
total_time, 1e-3)
gnn_feature[i, j, 5] = define.dis(resource[0], p_i[0], resource[1],
p_i[1]) / define.speed / max(
total_time, 1e-3)
gnn_feature[i, j, 6] = define.dis(resource[0], p_j[0], resource[1],
p_j[1]) / define.speed / max(
total_time, 1e-3)
for k in range(2, 7):
if gnn_feature[i, j, k] > 1:
tmp = gnn_feature[i, j, k] - 1
gnn_feature[i, j, k] = 2 - np.exp(-tmp)
mask[i, j] = 1
def getReturnTime(self, p):
end_x, end_y = self.__resource.getInitialPos()
p_x, p_y = p.getPosition()
t = define.dis(end_x, p_x, end_y, p_y) / self.speed
return t