def s(self):
"""
Формирование матрицы коэффициентов системы
:return:
"""
result = np.zeros(
[2 * self.M1 * self.K + self.M + self.M1 + 1, self.M1 * (2 * self.K - 1) + self.M + 2 * self.M1 + 1],
complex)
w = self.wf(0)
for n in xrange(0, self.M + 1):
for m in xrange(0, self.M + 1):
result[n, m] = delta(n, m)
result[n, m + self.M1] = - w[n, m]
result[n, m + 2 * self.M1] = - w[n, m]
result[n + self.M1, m] = - w.transpose()[n, m]
result[n + self.M1, m + self.M1] = - delta(n, m)
result[n + self.M1, m + 2 * self.M1] = delta(n, m)
if self.K > 1:
for j in xrange(1, self.K):
w = self.wf(j)
p1 = self.psi1(j)
p2 = self.psi2(j)
for n in xrange(0, self.M + 1):
for m in xrange(0, self.M + 1):
result[2 * self.M1 * j + n, 2 * self.M1 * j + m - self.M1] = p1[n, m]
result[2 * self.M1 * j + n, 2 * self.M1 * j + m + 0] = p2[n, m]
result[2 * self.M1 * j + n, 2 * self.M1 * j + m + 1 * self.M1] = - w[n, m]
result[2 * self.M1 * j + n, 2 * self.M1 * j + m + 2 * self.M1] = - w[n, m]
result[2 * self.M1 * j + n + self.M1, 2 * self.M1 * j + m - self.M1] = \
np.dot(w.transpose(), p1)[n, m]
result[2 * self.M1 * j + n + self.M1, 2 * self.M1 * j + m + 0] = np.dot(- w.transpose(), p2)[
n, m]
result[2 * self.M1 * j + n + self.M1, 2 * self.M1 * j + m + 1 * self.M1] = - delta(n, m)
result[2 * self.M1 * j + n + self.M1, 2 * self.M1 * j + m + 2 * self.M1] = delta(n, m)
w = self.wf(self.K)
p1 = self.psi1(self.K)
p2 = self.psi2(self.K)
for n in xrange(0, self.M + 1):
for m in xrange(0, self.M + 1):
result[2 * self.M1 * self.K + n, 1 * self.M1 * (2 * self.K - 1) + m - 0] = p1[n, m]
result[2 * self.M1 * self.K + n, 1 * self.M1 * (2 * self.K - 1) + m + 1 * self.M1] = p2[n, m]
result[2 * self.M1 * self.K + n, 1 * self.M1 * (2 * self.K - 1) + m + 2 * self.M1] = - w[n, m]
result[2 * self.M1 * self.K + n + self.M1, 1 * self.M1 * (2 * self.K - 1) + m - 0] = \
np.dot(w.transpose(), p1)[n, m]
result[2 * self.M1 * self.K + n + self.M1, 1 * self.M1 * (2 * self.K - 1) + m + 1 * self.M1] = \
np.dot(- w.transpose(), p2)[n, m]
result[2 * self.M1 * self.K + n + self.M1, 1 * self.M1 * (2 * self.K - 1) + m + 2 * self.M1] = - delta(
m, n)
return result
def getMaxReach(avePos):
distance = 0
for strut in STRUTS.values():
for end in strut:
dist = ut.delta(end, avePos)
if dist > distance:
distance = dist
return distance
def getDistances():
distances = []
for span in SPANS:
distances.append(ut.delta(span[0][1], span[1][1]))
global DISTANCES
DISTANCES.clear()
DISTANCES = distances
def calcula_topographic_error(mapa): erro = 0 for objeto in mapa.objetos: cluster1 = objeto.cluster cluster2 = objeto.segundo distancia = util.delta(cluster1.point, cluster2.point) if distancia > 2: erro += 1 tam = len(mapa.objetos) erro_topografico = erro/tam return erro_topografico
def energy_E2(V):
P_set, Q_set = util.PQ_N4(I, P)
V_P = V[P_set[0], P_set[1]]
V_Q = V[Q_set[0], Q_set[1]]
S_P = S[P_set[0], P_set[1]]
S_Q = S[Q_set[0], Q_set[1]]
H_P = H[P_set[0], P_set[1]]
H_Q = H[Q_set[0], Q_set[1]]
delta = util.delta(V_P, V_Q)
s_max = np.max((S_P, S_Q), axis=0)
d = np.deg2rad(util.deg_distance(H_P, H_Q))
e2 = np.multiply(np.multiply(delta, s_max), np.reciprocal(d))
#print(delta, s_max, np.multiply( delta, s_max ), np.reciprocal(d.astype(float)), e2)
e2 = np.sum(e2)
return e2
def energy_E2(self, X, V, P):
# Opencv store H as [0, 180) --> [0, 360)
H = X[:, :, 0].astype(np.int32)* 2
# Opencv store S as [0, 255] --> [0, 1]
S = X[:, :, 1].astype(np.float32) / 255.0
P_set, Q_set = PQ_N4(X, P)
V_P = V[ P_set[0], P_set[1] ]
V_Q = V[ Q_set[0], Q_set[1] ]
S_P = S[ P_set[0], P_set[1] ]
S_Q = S[ Q_set[0], Q_set[1] ]
H_P = H[ P_set[0], P_set[1] ]
H_Q = H[ Q_set[0], Q_set[1] ]
delta = util.delta( V_p, V_q )
s_max = np.max((S_P, S_Q), axis=0)
d = util.deg_distance(H_P, H_Q)
e2 = np.multiply( np.multiply( delta, s_max ), np.reciprocal(d) )
e2 = np.sum(e2)
return e2
def __repr__(self):
st = self.started and self.start_time
end = self.completed and self.end_time
s = 'Job {} [{} -> {} -> {} : run {} limit {} proc {}]'
return s.format(self.ID, delta(self.submit), delta(st), delta(end),
delta(self.run_time), delta(self.time_limit), self.proc)
def __repr__(self):
s = 'Camp {} {} [created {} work {} left {} : jobs {} {}]'
return s.format(self.ID, self.user.ID, delta(self.created),
delta(self.workload), delta(self.time_left),
len(self.active_jobs), len(self.completed_jobs))
def run(self):
"""
Proceed with the simulation.
Return a list of encountered events.
"""
# Magic value taken from slurm/multifactor plugin.
self._decay_factor = 1 - (0.693 / self._settings.decay)
# Note:
# The CPU usage decay is always applied after each event.
# There is also a dummy `force_decay` event inserted into
# the queue to force the calculations in case the gap
# between consecutive events would be too long.
# TODO ZAMIENIC APPLY DECAY NA BACKGROUND THREAD TAK SAMO JAK BACKFILLING?!?!?!
# TODO WTEDY BEDZIE TRZEBA TRZYMAC "TEMPORARY" USAGE GDZIES ODDZIELNIE I GO DODAWAC
# TODO DO GLOWNEGO DOPIERO PRZY URUCHOMIENIU TEGO "THREADA"
self._force_period = 60 * 5
self._initialize()
sub_iter = sub_count = 0
sub_total = len(self._block)
end_iter = 0
schedule = backfill = False
instant_bf = self._settings.bf_depth and not self._settings.bf_interval
# the first job submission is the simulation 'time zero'
prev_event = self._block[0].submit
self._diag.prev_util["time"] = prev_event
visual_update = 60 # notify the user about the progress
# TODO DODAC POZA CZASEM TEZ PROCENTOWO CO 25%
next_visual = time.time() + visual_update
while sub_iter < sub_total or not self._pq.empty():
# We only need to keep two `new_job` events in the
# queue at the same time (one to process, one to peek).
while sub_iter < sub_total and sub_count < 2:
self._pq.add(self._block[sub_iter].submit, Events.new_job, self._block[sub_iter])
sub_iter += 1
sub_count += 1
# the queue cannot be empty here
self._now, event, entity = self._pq.pop()
if event != Events.force_decay:
logging.debug("Time %s, event %s", delta(self._now), event)
# Process the time skipped between events
# before changing the state of the system.
diff = self._now - prev_event
if diff:
self._virt_first_stage(diff)
self._real_first_stage(diff)
# The default flow is to redistribute the virtual
# time and compute new campaign ends (and maybe do
# a scheduling / backfilling pass in the between).
virt_second = True
campaigns = True
if event == Events.new_job:
# check if the job is runnable
if self._manager.runnable(entity):
self._new_job_event(entity)
schedule = True
else:
self._diag.skipped += 1
end_iter += 1
sub_count -= 1
elif event == Events.job_end:
self._job_end_event(entity)
end_iter += 1
schedule = True
elif event == Events.estimate_end:
self._estimate_end_event(entity)
elif event == Events.bf_run:
backfill = True
elif event == Events.campaign_end:
# We need to redistribute the virtual time now,
# so the campaign can actually end.
self._virt_second_stage()
virt_second = False # already done
campaigns = self._camp_end_event(entity)
elif event == Events.force_decay:
self._diag.forced += 1
virt_second = False # no need to do it now
campaigns = False # no change to campaign ends
else:
raise Exception("unknown event")
# update event timer
prev_event = self._now
if not self._pq.empty():
# We need to process all the events that happen at
# the same time *AND* change the campaign workloads
# before we can continue further.
next_time, next_event, _ = self._pq.peek()
if next_time == self._now and next_event < Events.bf_run:
continue
if virt_second:
self._virt_second_stage()
if schedule:
scheduled_jobs = self._schedule(bf_mode=False)
self._diag.sched_jobs += scheduled_jobs
self._diag.sched_pass += 1
self._update_util() # must be after schedule
schedule = False
if instant_bf:
backfill = True
if backfill:
backfilled_jobs = self._schedule(bf_mode=True)
self._diag.bf_jobs += backfilled_jobs
self._diag.bf_pass += 1
self._update_util() # must be after schedule
backfill = False
if campaigns:
self._update_camp_estimates()
# add periodically occurring events
if event < Events.bf_run:
self._next_backfill(self._now)
elif event == Events.bf_run and backfilled_jobs:
self._next_backfill(self._now + 1)
if end_iter < sub_total:
# There are still jobs in the simulation
# so we need an accurate usage.
assert not self._pq.empty(), "infinite loop"
self._next_force_decay()
# progress report
if time.time() > next_visual:
next_visual += visual_update
comp = float(sub_iter + end_iter) / (2 * sub_total)
msg = "Block {:2} scheduler {}: {} completed {:.2f}%"
logging.info(
msg.format(self._block.number, self._parts.scheduler, time.strftime("%H:%M:%S"), comp * 100)
)
if self._waiting_jobs:
top_prio = self._waiting_jobs[-1].proc
else:
top_prio = -1
logging.info(
"events {} {} | cpus {} {} | waiting jobs {} {}"
" | stats {:.2f} {:.2f} {:.2f}".format(
sub_iter,
end_iter,
self._stats.cpu_used,
self._cpu_free,
len(self._waiting_jobs),
top_prio,
(self._diag.avg_util["sum"] / self._diag.avg_util["period"]),
self._diag.sched_jobs / float(sub_iter),
self._diag.bf_jobs / float(sub_iter),
)
)
self._finalize()
# Results for each user should be in this order:
# 1) job ends (this is done during simulation)
# 2) camp ends
# 3) user stats
for u in self._users.itervalues():
for i, c in enumerate(u.completed_camps):
self._store_camp_ended(c)
self._store_user_stats(u)
# merge the results
self._results.append(self._compressor.flush())
self._results = "".join(self._results)
return self._results, self._diag
def __repr__(self):
s = '[{}, {}] last {} first {}\n\tavail {}\n\trsrvd {}'
return s.format(delta(self.begin), delta(self.end),
self.job_ends, self.rsrv_starts,
self.avail, self.reserved)
def run(self):
"""
Proceed with the simulation.
Return a list of encountered events.
"""
# Magic value taken from slurm/multifactor plugin.
self._decay_factor = 1 - (0.693 / self._settings.decay)
# Note:
# The CPU usage decay is always applied after each event.
# There is also a dummy `force_decay` event inserted into
# the queue to force the calculations in case the gap
# between consecutive events would be too long.
self._force_period = 60 * 5
self._initialize()
sub_iter = sub_count = 0
sub_total = len(self._block)
end_iter = 0
schedule = backfill = False
instant_bf = (self._settings.bf_depth and
not self._settings.bf_interval)
# the first job submission is the simulation 'time zero'
prev_event = self._block[0].submit
self._diag.prev_util['time'] = prev_event
# time to notify the user about the simulation progress
next_visual_update = time.time() + self._settings.update_time
while sub_iter < sub_total or not self._pq.empty():
# We only need to keep two `new_job` events in the
# queue at the same time (one to process, one to peek).
while sub_iter < sub_total and sub_count < 2:
self._pq.add(
self._block[sub_iter].submit,
Events.new_job,
self._block[sub_iter]
)
sub_iter += 1
sub_count += 1
# the queue cannot be empty here
self._now, event, entity = self._pq.pop()
if event != Events.force_decay:
logging.debug('Time %s, event %s', delta(self._now), event)
# Process the time skipped between events
# before changing the state of the system.
diff = self._now - prev_event
if diff:
self._virt_first_stage(diff)
self._real_first_stage(diff)
# The default flow is to redistribute the virtual
# time and compute new campaign ends (and maybe do
# a scheduling / backfilling pass in the between).
virt_second = True
campaigns = True
if event == Events.new_job:
# check if the job is runnable
if self._manager.runnable(entity):
self._new_job_event(entity)
schedule = True
else:
self._diag.skipped += 1
end_iter += 1
sub_count -= 1
elif event == Events.job_end:
self._job_end_event(entity)
end_iter += 1
schedule = True
elif event == Events.estimate_end:
self._estimate_end_event(entity)
elif event == Events.bf_run:
backfill = True
elif event == Events.campaign_end:
# We need to redistribute the virtual time now,
# so the campaign can actually end.
self._virt_second_stage()
virt_second = False # already done
campaigns = self._camp_end_event(entity)
elif event == Events.force_decay:
self._diag.forced += 1
virt_second = False # no need to do it now
campaigns = False # no change to campaign ends
else:
raise Exception('unknown event')
# update event timer
prev_event = self._now
if not self._pq.empty():
# We need to process all the events that happen at
# the same time *AND* change the campaign workloads
# before we can continue further.
next_time, next_event, _ = self._pq.peek()
if (next_time == self._now and
next_event < Events.bf_run):
continue
if virt_second:
self._virt_second_stage()
if schedule:
scheduled_jobs = self._schedule(bf_mode=False)
self._diag.sched_jobs += scheduled_jobs
self._diag.sched_pass += 1
self._update_util() # must be after schedule
schedule = False
if instant_bf: backfill = True
if backfill:
backfilled_jobs = self._schedule(bf_mode=True)
self._diag.bf_jobs += backfilled_jobs
self._diag.bf_pass += 1
self._update_util() # must be after schedule
backfill = False
if campaigns:
self._update_camp_estimates()
# add periodically occurring events
if event < Events.bf_run:
self._next_backfill(self._now)
elif event == Events.bf_run and backfilled_jobs:
self._next_backfill(self._now + 1)
if end_iter < sub_total:
# There are still jobs in the simulation
# so we need an accurate usage.
assert not self._pq.empty(), 'infinite loop'
self._next_force_decay()
# progress report
if time.time() > next_visual_update:
completed = float(sub_iter + end_iter) / (2 * sub_total)
self._log_progress(sub_iter, completed)
next_visual_update += self._settings.update_time
self._finalize()
# Results for each user should be in this order:
# 1) job ends (this is done during simulation)
# 2) camp ends
# 3) user stats
for u in self._users.itervalues():
for i, c in enumerate(u.completed_camps):
self._store_camp_ended(c)
self._store_user_stats(u)
# merge the results
self._results.append(self._compressor.flush())
self._results = ''.join(self._results)
return self._results, self._diag
def psi2(self, num):
"""
:param num:
:return:
"""
result = np.zeros([self.M + 1, self.M + 1], complex)
for n in xrange(0, self.M + 1):
for m in xrange(0, self.M + 1):
result[n, m] = (sp.e ** ((0 + 1j) * self.beta(num, n) * self.deltaL)) * delta(n, m)
return result