def ExactS(traj_c, traj_q):
subsim = 999999
subtraj = [0, len(traj_c) - 1]
subset = {}
N = len(traj_c)
for i in range(N):
DIS = Distance(len(traj_c[i:]), len(traj_q))
for j in range(i, N):
temp = DIS.FRECHET(traj_c[i:j + 1], traj_q)
#temp = similaritymeasures.frechet(traj_c[i:j+1], traj_q)[0]
#print('sub-range:', [i, j], temp)
subset[(i, j)] = temp
if temp < subsim:
subsim = temp
subtraj = [i, j]
return subsim, subtraj, subset
def SizeS(traj_c, traj_q, par=5):
L = len(traj_q)
L_lo = min(len(traj_c), int((L - par)))
L_up = min(len(traj_c), int((L + par)))
subsim = 999999
subtraj = [0, len(traj_c) - 1]
N = len(traj_c)
for i in range(N):
DIS = Distance(len(traj_c[i:]), len(traj_q))
for j in range(i, N):
if (j - i + 1) < L_lo or (j - i + 1) > L_up:
continue
temp = DIS.FRECHET(traj_c[i:j + 1], traj_q)
#temp = similaritymeasures.FRECHET(traj_c[i:j+1], traj_q)[0]
#print('sub-range:', [i, j], temp)
if temp < subsim:
subsim = temp
subtraj = [i, j]
return subsim, subtraj
class Subtraj():
def __init__(self, cand_train, query_train):
self.action_space = ['0', '1', '2', '3', '4']
self.n_actions = len(self.action_space)
self.n_features = 3
self.cand_train_name = cand_train
self.query_train_name = query_train
self.presim = 0
self.sufsim = 0
self.RW = 0.0
self._load()
def _load(self):
self.cand_train_data = pickle.load(open(self.cand_train_name, 'rb'),
encoding='bytes')
self.query_train_data = pickle.load(open(self.query_train_name, 'rb'),
encoding='bytes')
def reset(self, episode, label='E'):
# prefix_state --> [split_point, index]
# suffix_state --> [index + 1, len - 1]
# return observation
self.split_point = 0
self.DIS = Distance(len(self.cand_train_data[episode]),
len(self.query_train_data[episode]))
self.DIS_R = Distance(len(self.cand_train_data[episode]),
len(self.query_train_data[episode]))
self.length = len(self.cand_train_data[episode])
self.skip = []
self.presim = self.DIS.FRECHET(
self.cand_train_data[episode][self.split_point:1],
self.query_train_data[episode])
whole = self.DIS_R.FRECHET(self.cand_train_data[episode][::-1],
self.query_train_data[episode][::-1])
self.sufsim = self.DIS_R.D[
self.length - 2,
-1] #self.DIS_R.FRECHET(self.cand_train_data[episode][1:][::-1],self.query_train_data[episode][::-1])
#print('reset',self.sufsim,self.DIS_R.D[self.length-2,-1])
#self.DIS.FRECHET(self.cand_train_data[episode], self.query_train_data[episode])
observation = np.array([whole, self.presim,
self.sufsim]).reshape(1, -1)
self.subsim = min(whole, self.presim, self.sufsim)
#print('episode', episode, whole, self.presim, self.sufsim)
if self.subsim == whole:
self.subtraj = [0, self.length - 1]
if self.subsim == self.presim:
self.subtraj = [0, 0]
if self.subsim == self.sufsim:
self.subtraj = [1, self.length - 1]
if label == 'T':
self.REWARD_DIS = Distance(len(self.cand_train_data[episode]),
len(self.query_train_data[episode]))
self.presim_real = self.REWARD_DIS.FRECHET(
self.cand_train_data[episode][self.split_point:1],
self.query_train_data[episode])
self.subsim_real = self.subsim
return observation, self.length, -1
def step(self, episode, action, index, label='E'):
if action == 0: #non-split
#state transfer
if index == self.length - 1:
done = True
else:
done = False
self.presim = self.DIS.FRECHET(
self.cand_train_data[episode][self.split_point:(index + 1)],
self.query_train_data[episode], self.skip)
self.sufsim = self.DIS_R.D[
self.length - 2 - index,
-1] #self.DIS_R.FRECHET(self.cand_train_data[episode][(index+1):][::-1],self.query_train_data[episode][::-1])
#print('A0', self.sufsim, self.DIS_R.D[self.length-2-index,-1])
if (index + 1) == self.length:
self.sufsim = self.presim
observation = np.array([self.subsim, self.presim,
self.sufsim]).reshape(1, -1)
if self.presim < self.subsim:
self.subsim = self.presim
self.subtraj = [self.split_point, index]
if self.sufsim < self.subsim:
self.subsim = self.sufsim
self.subtraj = [index + 1, self.length - 1]
if label == 'T':
last_subsim = self.subsim_real
self.presim_real = self.REWARD_DIS.FRECHET(
self.cand_train_data[episode][self.split_point:(index +
1)],
self.query_train_data[episode])
self.subsim_real = min(self.presim_real, self.sufsim,
last_subsim)
self.RW = last_subsim - self.subsim_real
#print('action0', self.RW)
#print(self.presim, self.presim_real)
return observation, self.RW, done, -1
if action == 1: #split
if index == self.length - 1:
done = True
else:
done = False
self.skip = []
self.split_point = index
self.DIS = Distance(
len(self.cand_train_data[episode][self.split_point:]),
len(self.query_train_data[episode]))
#state transfer
self.presim = self.DIS.FRECHET(
self.cand_train_data[episode][self.split_point:(index + 1)],
self.query_train_data[episode], self.skip)
self.sufsim = self.DIS_R.D[
self.length - 2 - index,
-1] #self.DIS_R.FRECHET(self.cand_train_data[episode][(index+1):][::-1],self.query_train_data[episode][::-1])
#print('A1', self.sufsim, self.DIS_R.D[self.length-2-index,-1])
if (index + 1) == self.length:
self.sufsim = self.presim
observation = np.array([self.subsim, self.presim,
self.sufsim]).reshape(1, -1)
if self.presim < self.subsim:
self.subsim = self.presim
self.subtraj = [self.split_point, index]
if self.sufsim < self.subsim:
self.subsim = self.sufsim
self.subtraj = [index + 1, self.length - 1]
if label == 'T':
last_subsim = self.subsim_real
self.REWARD_DIS = Distance(
len(self.cand_train_data[episode][self.split_point:]),
len(self.query_train_data[episode]))
self.presim_real = self.REWARD_DIS.FRECHET(
self.cand_train_data[episode][self.split_point:(index +
1)],
self.query_train_data[episode])
self.subsim_real = min(self.presim_real, self.sufsim,
last_subsim)
self.RW = last_subsim - self.subsim_real
#print('action1', self.RW)
#print(self.presim, self.presim_real)
return observation, self.RW, done, -1
if action > 1: #skip
INX = min(index + action - 1, self.length - 1)
if INX == self.length - 1:
done = True
else:
done = False
for i in range(index, INX):
self.skip.append(list(self.cand_train_data[episode][i]))
self.presim = self.DIS.FRECHET(
self.cand_train_data[episode][self.split_point:(INX + 1)],
self.query_train_data[episode], self.skip)
self.sufsim = self.DIS_R.D[
self.length - 2 - INX,
-1] #self.DIS_R.FRECHET(self.cand_train_data[episode][(INX+1):][::-1],self.query_train_data[episode][::-1])
#print('A2', self.sufsim, self.DIS_R.D[self.length-2-INX,-1])
if (INX + 1) == self.length:
self.sufsim = self.presim
observation = np.array([self.subsim, self.presim,
self.sufsim]).reshape(1, -1)
if self.presim < self.subsim:
self.subsim = self.presim
self.subtraj = [self.split_point, INX]
if self.sufsim < self.subsim:
self.subsim = self.sufsim
self.subtraj = [INX + 1, self.length - 1]
if label == 'T':
last_subsim = self.subsim_real
self.presim_real = self.REWARD_DIS.FRECHET(
self.cand_train_data[episode][self.split_point:(INX + 1)],
self.query_train_data[episode])
self.subsim_real = min(self.presim_real, self.sufsim,
last_subsim)
self.RW = last_subsim - self.subsim_real
#print(self.presim, self.presim_real)
return observation, self.RW, done, INX
def output(self, index, episode, label='E'):
#print('check', self.subsim, self.subtraj)
if label == 'T':
#print('check', self.subsim, self.subtraj, self.subsim_real)
return [self.subsim_real, self.subtraj]
if label == 'E':
self.DIS = Distance(
len(self.cand_train_data[episode]
[self.subtraj[0]:self.subtraj[1] + 1]),
len(self.query_train_data[episode]))
self.subsim_real = self.DIS.FRECHET(
self.cand_train_data[episode][self.subtraj[0]:self.subtraj[1] +
1],
self.query_train_data[episode])
return [self.subsim_real, self.subtraj]
class Subtraj():
def __init__(self, cand_train, query_train):
self.action_space = ['0', '1']
self.n_actions = len(self.action_space)
self.n_features = 3
self.cand_train_name = cand_train
self.query_train_name = query_train
self.presim = 0
self.sufsim = 0
self.RW = 0.0
self._load()
def _load(self):
self.cand_train_data = pickle.load(open(self.cand_train_name, 'rb'), encoding='bytes')
self.query_train_data = pickle.load(open(self.query_train_name, 'rb'), encoding='bytes')
def reset(self, episode):
# prefix_state --> [split_point, index]
# suffix_state --> [index + 1, len - 1]
# return observation
self.split_point = 0
self.DIS = Distance(len(self.cand_train_data[episode]), len(self.query_train_data[episode]))
self.DIS_R = Distance(len(self.cand_train_data[episode]), len(self.query_train_data[episode]))
self.length = len(self.cand_train_data[episode])
self.presim = self.DIS.FRECHET(self.cand_train_data[episode][self.split_point:1], self.query_train_data[episode])
self.sufsim = self.DIS_R.FRECHET(self.cand_train_data[episode][1:][::-1],self.query_train_data[episode][::-1])
whole = self.DIS_R.FRECHET(self.cand_train_data[episode][::-1],self.query_train_data[episode][::-1]) #self.DIS.FRECHET(self.cand_train_data[episode], self.query_train_data[episode])
observation = np.array([whole, self.presim, self.sufsim]).reshape(1,-1)
self.subsim = min(whole, self.presim, self.sufsim)
#print('episode', episode, whole, self.presim, self.sufsim)
if self.subsim == whole:
self.subtraj = [0, self.length - 1]
if self.subsim == self.presim:
self.subtraj = [0, 0]
if self.subsim == self.sufsim:
self.subtraj = [1, self.length - 1]
return observation, self.length
def step(self, episode, action, index):
if action == 0: #non-split
#state transfer
self.presim = self.DIS.FRECHET(self.cand_train_data[episode][self.split_point:(index+1)], self.query_train_data[episode])
self.sufsim = self.DIS_R.FRECHET(self.cand_train_data[episode][(index+1):][::-1],self.query_train_data[episode][::-1])
if (index+1) == self.length:
self.sufsim = self.presim
observation = np.array([self.subsim, self.presim, self.sufsim]).reshape(1,-1)
last_subsim = self.subsim
if self.presim < self.subsim:
self.subsim = self.presim
self.subtraj = [self.split_point, index]
if self.sufsim < self.subsim:
self.subsim = self.sufsim
self.subtraj = [index + 1, self.length - 1]
self.RW = last_subsim - self.subsim
#print('action0', self.RW)
return observation, self.RW
if action == 1: #split
self.split_point = index
self.DIS = Distance(len(self.cand_train_data[episode][self.split_point:]), len(self.query_train_data[episode]))
#state transfer
self.presim = self.DIS.FRECHET(self.cand_train_data[episode][self.split_point:(index+1)], self.query_train_data[episode])
self.sufsim = self.DIS_R.FRECHET(self.cand_train_data[episode][(index+1):][::-1],self.query_train_data[episode][::-1])
if (index+1) == self.length:
self.sufsim = self.presim
observation = np.array([self.subsim, self.presim, self.sufsim]).reshape(1,-1)
last_subsim = self.subsim
if self.presim < self.subsim:
self.subsim = self.presim
self.subtraj = [self.split_point, index]
if self.sufsim < self.subsim:
self.subsim = self.sufsim
self.subtraj = [index + 1, self.length - 1]
self.RW = last_subsim - self.subsim
#print('action1', self.RW)
return observation, self.RW
def output(self, index, episode):
#print('check', self.subsim, self.subtraj)
return [self.subsim, self.subtraj]
def heuristic(traj_c, traj_q, opt, delay_K=5):
delay = 0
subsim = 999999
subtraj = [0, len(traj_c)-1]
split_point = 0
DIS = Distance(len(traj_c), len(traj_q))
DIS_R = Distance(len(traj_c), len(traj_q))
pos_d_coll = []
pos_d_f = False
temp = 'non'
if opt != 'POS-D':
for i in range(len(traj_c)):
#submit prefix
presim = DIS.FRECHET(traj_c[split_point:i+1],traj_q)
sufsim = heuristic_suffix_opt(traj_c, traj_q, i+1, opt, DIS_R)
#print('-> maintain:', subtraj, subsim)
#print('prefix:', [split_point, i], presim)
#print('suffix:', [i+1, len(traj_c)-1], sufsim)S
if presim < subsim or sufsim < subsim:
temp = i + 1
subsim = min(presim, sufsim)
if presim < sufsim:
subtraj = [split_point, (temp-1)]
else:
subtraj = [temp, len(traj_c)-1]
split_point = temp
DIS = Distance(len(traj_c[i:]), len(traj_q))
else:
i = -1
while True:
i = i + 1
if temp != 'non':
i = temp
temp = 'non'
if i == len(traj_c) - 1:
break
#submit prefix
presim = DIS.FRECHET(traj_c[split_point:i+1],traj_q)
if pos_d_f == False and presim < subsim: #open delay
pos_d_f = True
if pos_d_f == True and delay < delay_K:
delay = delay + 1
pos_d_coll.append((presim, i))
continue
if pos_d_f == True and delay == delay_K:
sort = sorted(pos_d_coll, key=lambda d: d[0])
temp = sort[0][1] + 1
subsim = sort[0][0]
subtraj = [split_point, (temp-1)]
split_point = temp
DIS = Distance(len(traj_c[sort[0][1]:]), len(traj_q))
delay = 0
pos_d_f = False
pos_d_coll = []
if subsim == 999999: #for extreme cases
if pos_d_coll == []:
presim = DIS.FRECHET(traj_c[split_point:i+1],traj_q)
pos_d_coll.append((presim, i))
sort = sorted(pos_d_coll, key=lambda d: d[0])
temp = sort[0][1] + 1
subsim = sort[0][0]
subtraj = [split_point, (temp-1)]
return subsim, subtraj