def findReplacement(speaker, date, specialization, piH, schedule):
for dateA in sorted(schedule):
if dateA <= date:
continue
for i, speakerA in enumerate(schedule[dateA]):
if speakerA == speaker:
continue
slot = '%s,%d' % (dateA, i)
spec = g_.node[slot]['specialization']
if spec != specialization:
continue
thisPI = speakers_[speakerA]['pi_or_host']
if thisPI == piH:
continue
# final check. Make sure the AWS does not come too early.
nDays = (toDate(dateA) - toDate(date)).days
prevAWS = g_.node[speakerA]['last_date']
newDate = prevAWS - datetime.timedelta(days=nDays)
if (newDate - prevAWS).days < 400:
_logger.warn('Costly swapping. ignoring...')
continue
return speakerA, dateA
_logger.warn('Could not find replacement for %s.%s' % (speaker, date))
return None
def main(outfile):
global db_
_logger.info('Scheduling AWS')
getAllAWSPlusUpcoming()
ans = None
try:
construct_flow_graph()
ans = computeSchedule()
except Exception as e:
_logger.warn("Failed to schedule. Error was %s" % e)
try:
print_schedule(ans, outfile)
except Exception as e:
_logger.error("Could not print schedule. %s" % e)
if ans:
commit_schedule(ans)
else:
print('Failed to compute schedule')
return -1
try:
write_graph()
except Exception as e:
_logger.error("Could not write graph to file")
_logger.error("\tError was %s" % e)
db_.close()
def main(outfile):
global db_
_logger.info('Scheduling AWS')
getAllAWSPlusUpcoming()
ans = None
try:
construct_flow_graph()
ans = computeSchedule()
except Exception as e:
_logger.warn("Failed to schedule. Error was %s" % e)
# update specialization on g_ nodes.
for date in ans:
specs = [specialization_.get(x, 'UNSPECIFIED') for x in ans[date]]
mostCommonSpec, freq = Counter(specs).most_common(1)[0]
for slot in range(3):
slot = '%s,%d' % (date, slot)
if slot in g_:
g_.node[slot]['specialization'] = mostCommonSpec
ans = group_schedule(ans)
ans = aws_helper.no_common_labs(ans)
if ans:
commit_schedule(ans)
else:
print('Failed to compute/commit schedule')
return -1
try:
print_schedule(ans, outfile)
except Exception as e:
_logger.error("Could not print schedule. %s" % e)
try:
write_graph()
except Exception as e:
_logger.error("Could not write graph to file")
_logger.error("\tError was %s" % e)
db_.close()
def no_common_labs(schedule, nweeks=2, ncalls=0):
# Make sure that first 2 week entries have different PIs.
if ncalls > 100:
_logger.warn("Terminated after 100 calls")
return schedule
failedDates = []
sortedDates = sorted(schedule)
for ix, date in enumerate(sortedDates[:nweeks]):
labs = []
for i, speaker in enumerate(schedule[date]):
spec = g_.node['%s,%d' % (date, i)]['specialization']
piH = speakers_[speaker]['pi_or_host']
if piH in labs:
spec = specialization_.get(speaker, 'UNSPECIFIED')
replaceWith = findReplacement(speaker, date, spec, piH,
schedule)
if replaceWith is not None:
speakerB, dateB = replaceWith
speakerA, dateA = speaker, date
schedule[dateA].append(speakerB)
schedule[dateA].remove(speakerA)
schedule[dateB].append(speakerA)
schedule[dateB].remove(speakerB)
_logger.info('Swapping %s and %s' % (speakerA, speakerB))
else:
# swap this row by next and try again.
failedDates.append(date)
_logger.info("Failed to find alternative for %s" % date)
# swap this date with someone else.
for iy, datey in enumerate(sortedDates[nweeks:]):
if len(schedule[datey]) == len(schedule[date]):
# we can swap with entry.
temp = schedule[datey]
schedule[datey] = schedule[date]
schedule[date] = temp
return no_common_labs(schedule, nweeks, ncalls + 1)
else:
labs.append(piH)
for fd in failedDates:
_logger.warn('Entry for date %s has multiple speakers from same lab' %
fd)
_logger.warn('Moving whole row down to more than %d positions' %
nweeks)
return schedule
def construct_flow_graph():
"""This is the most critical section of this task. It is usually good if
flow graph is constructed to honor policy as much as possible rather than
fixing the solution later.
One important scheduling aim is to minimize number of freshers on the same
day. Ideally no more than 1 freshers should be allowed on same day. This can
be achieved by modifying the solution later : swapping freshers with
experienced speaker from other days. We avoid that by drawing two edges
from freshers to a 'date' i.e. maximum of 2 slots can be filled by freshers.
For others we let them fill all three slots.
"""
g_.add_node('source', pos=(0, 0))
g_.add_node('sink', pos=(10, 10))
lastDate = None
freshers = set()
for i, speaker in enumerate(speakers_):
# Last entry is most recent
if speaker not in aws_.keys():
# We are here because this speaker has not given any AWS yet.
freshers.add(speaker)
# If this user has PHD/POSTDOC, or INTPHD title. We create a dummy
# last date to bring her into potential speakers.
# First make sure, I have their date of joining. Otherwise I
# can't continue. For MSc By Research/INTPHD assign their first AWS
# after 18 months. For PHD and POSTDOC, it should be after 12 months.
if speakers_[speaker]['title'] == 'INTPHD':
# InPhd should get their first AWS after 15 months of
# joining.
_logger.info('%s = INTPHD with 0 AWS so far' % speaker)
joinDate = speakers_[speaker]['joined_on']
if not joinDate:
_logger.warn("Could not find joining date")
else:
lastDate = monthdelta(joinDate, +6)
if speakers_[speaker]['title'] == 'MSC':
# MSc should get their first AWS after 18 months of
# joining. Same as INTPHD
_logger.info('%s = MSC BY RESEARCH with 0 AWS so far' %
speaker)
joinDate = speakers_[speaker]['joined_on']
if not joinDate:
_logger.warn("Could not find joining date")
else:
lastDate = monthdelta(joinDate, +6)
elif speakers_[speaker]['title'] in ['PHD', 'POSTDOC']:
joinDate = speakers_[speaker]['joined_on']
_logger.info('%s PHD/POSTDOC with 0 AWS so far' % speaker)
if not joinDate:
_logger.warn("Could not find joining date")
else:
try:
# if datetime.
lastDate = joinDate.date()
except Exception as e:
# Else its date
lastDate = joinDate
else:
# We are here because this speaker has given AWS before
# If this speaker is already on upcoming AWS list, ignore it.
if speaker in upcoming_aws_:
_logger.info('Speaker %s is already scheduled on %s' %
(speaker, upcoming_aws_[speaker]))
continue
# If this speakers is MSC by research and has given AWS before, she/he
# need not give another.
elif speakers_[speaker]['title'] == 'MSC':
_logger.info('%s is MSC and has given AWS in the past' %
speaker)
continue
else:
lastDate = aws_[speaker][-1]['date']
# If a speaker has a lastDate either because he has given AWS in the
# past or becuase she is fresher. Create an edge.
if lastDate:
g_.add_node(speaker, last_date=lastDate, pos=(1, 3 * i))
g_.add_edge('source', speaker, capacity=1, weight=0)
# Compute totalWeeks of schedule starting today.
totalWeeks = int(365 / 7.0)
today = datetime.date.today()
nextMonday = today + datetime.timedelta(days=-today.weekday(), weeks=1)
slots = []
_logger.info("Computing for total %d weeks" % totalWeeks)
for i in range(totalWeeks):
nDays = i * 7
monday = nextMonday + datetime.timedelta(nDays)
# AWS don't care about holidays.
if monday in holidays_:
_logger.warn("This date %s is holiday" % monday)
continue
nSlots = 3
if monday in upcoming_aws_slots_:
# Check how many of these dates have been taken.
_logger.info('Date %s is taken ' % monday)
nSlots -= len(upcoming_aws_slots_[monday])
# For each Monday, we have 3 AWS - (assigned on upcoming_aws_slots_)
for j in range(nSlots):
dateSlot = '%s,%d' % (monday, j)
g_.add_node(dateSlot, date=monday, pos=(5, 10 * (3 * i + j)))
g_.add_edge(dateSlot, 'sink', capacity=1, weight=0)
slots.append(dateSlot)
# Now for each student, add potential edges.
idealGap = 7 * 60
# Keep edges from freshers to dates here. We allow maximum of 2 out of 3
# slots to be taken by freshers (maximum ).
freshersDate = defaultdict(list)
for speaker in speakers_:
preferences = aws_scheduling_requests_.get(speaker, {})
if preferences:
_logger.info("%s has preferences %s " % (speaker, preferences))
if speaker not in g_.nodes():
_logger.info('Nothing for user %s' % speaker)
continue
prevAWSDate = g_.node[speaker]['last_date']
for slot in slots:
date = g_.node[slot]['date']
weight = computeCost(speaker, date, prevAWSDate)
if weight:
# If the speaker is fresher, do not draw edges to all three
# slots. Draw just one but make sure that they get this slot. We
# reduce the cost to almost zero.
if speaker in freshers:
# Let two freshers take maximum of two slots on same day.
# The weight should be low but not lower than user
# preference.
if freshersDate.get(speaker, []).count(date) < 2:
addEdge(speaker, slot, 1, 5)
# This date is taken by this fresher.
freshersDate[speaker].append(date)
else:
addEdge(speaker, slot, 1, weight)
# Honour user preferences..
if preferences:
first = preferences.get('first_preference', None)
second = preferences.get('second_preference', None)
if first:
ndays = diffInDays(date, first, True)
if ndays <= 14:
_logger.info('Using first preference for %s' %
speaker)
addEdge(speaker, slot, 1, 0 + ndays / 7)
if second:
ndays = diffInDays(date, second, True)
if ndays <= 14:
_logger.info('Using second preference for %s' %
speaker)
addEdge(speaker, slot, 1, 2 + ndays / 7)
_logger.info('Constructed flow graph')
def construct_flow_graph(seed=0):
"""This is the most critical section of this task. It is usually good if
flow graph is constructed to honor policy as much as possible rather than
fixing the solution later.
One important scheduling aim is to minimize number of freshers on the same
day. Ideally no more than 1 freshers should be allowed on same day. This can
be achieved by modifying the solution later : swapping freshers with
experienced speaker from other days. We avoid that by drawing two edges
from freshers to a 'date' i.e. maximum of 2 slots can be filled by freshers.
For others we let them fill all three slots.
4 slots every monday and all belong to one specialization.
"""
g_.clear()
g_.add_node('source', pos=(0, 0))
g_.add_node('sink', pos=(10, 10))
# Compute totalWeeks of schedule starting today.
totalWeeks = 32
d = datetime.date.today()
while d.weekday() != 0: # 0 is monday
d += datetime.timedelta(days=1)
nextMonday = d
slots = []
weeks = [afterNDays(nextMonday, 7 * i) for i in range(totalWeeks)]
specializations = chooseSpecialization(totalWeeks, seed)
# Ignore the already filled upcoming AWS slots.
freq = Counter(speakersSpecialization_.values())
for monday in sorted(upcoming_aws_slots_):
speakers = upcoming_aws_slots_[monday]
for speaker in speakers:
specialization = speakersSpecialization_.get(
speaker, 'UNSPECIFIED')
freq[specialization] = max(0, freq[specialization] - 1)
# Update the frequencies.
for k in freq:
specializationFreqs_[k] = 1.0 * freq[k] / sum(freq.values())
# Collect all the valid slots with specialization in a list.
validSlots = []
for specForWeek, monday in zip(specializations, weeks):
# AWS don't care about holidays.
if monday in holidays_:
_logger.warn("This date %s is holiday" % monday)
continue
nSlots = 3
if monday in upcoming_aws_slots_:
# Check how many of these dates have been taken.
_logger.info('Date %s is taken ' % monday)
nSlots -= len(upcoming_aws_slots_[monday])
for i in range(nSlots):
validSlots.append(('%s,%s' % (monday, i), monday, specForWeek))
else:
validSlots += [('%s,%s' % (monday, i), monday, specForWeek)
for i in range(nSlots)]
# Keep edges from freshers to dates here. We allow maximum of 2 out of 3
# slots to be taken by freshers (maximum ).
missed = construct_graph(validSlots)
return missed
def construct_graph(validSlots):
lastDate = None
for i, speaker in enumerate(speakers_):
lastDate = get_prev_aws_date(speaker)
# If a speaker has a lastDate either because he has given AWS in the
# past or becuase she is fresher. Create an edge.
if lastDate is not None:
g_.add_node(speaker, last_date=lastDate, pos=(1, 3 * i))
g_.add_edge('source', speaker, capacity=1, weight=0)
for sid, monday, specForWeek in validSlots:
# For each Monday, we have 3 AWS - (assigned on upcoming_aws_slots_)
# For each week select a specialization.
_logger.info("++ Specialization for this week is %s" % specForWeek)
g_.add_node(sid, date=monday, specialization=specForWeek)
g_.add_edge(sid, 'sink', capacity=1, weight=0)
# Now for each student, add potential edges.
idealGap = 357
freshersDate = defaultdict(list)
for speaker in speakers_:
speakerSpecialization = speakersSpecialization_.get(speaker, '')
preferences = aws_scheduling_requests_.get(speaker, {})
if preferences:
_logger.info("%s has preferences %s " % (speaker, preferences))
if speaker not in g_.nodes():
_logger.info('Nothing for user %s' % speaker)
continue
prevAWSDate = g_.node[speaker]['last_date']
for slot, monday, speci in validSlots:
# If this slot does not belong to some specialization then ignore
# it.
if g_.node[slot]['specialization'] != speakerSpecialization:
continue
date = g_.node[slot]['date']
weight = computeCost(speaker, date, prevAWSDate)
if weight is None:
continue
# If the speaker is fresher, do not draw edges to all three
# slots. Draw just one but make sure that they get this slot. We
# reduce the cost to almost zero.
if speaker in freshers_:
# Let two freshers take maximum of two slots on same day.
# The weight should be low but not lower than user
# preference.
if freshersDate.get(speaker, []).count(date) < 2:
addEdge(speaker, slot, 1, 5)
# This date is taken by this fresher.
freshersDate[speaker].append(date)
else:
addEdge(speaker, slot, 1, weight)
# Honour user preferences..
if preferences:
first = preferences.get('first_preference', None)
second = preferences.get('second_preference', None)
if first:
ndays = diffInDays(date, first, True)
if ndays <= 14:
_logger.debug('Using first preference for %s' %
speaker)
addEdge(speaker, slot, 1, 0 + ndays / 7)
if second:
ndays = diffInDays(date, second, True)
if ndays <= 14:
_logger.info('Using second preference for %s' %
speaker)
addEdge(speaker, slot, 1, 2 + ndays / 7)
# Each slot node must have at least 3 nodes.
missedSlots = []
for slot, monday, speci in validSlots:
inDegree = g_.in_degree(slot)
inedges = g_.predecessors(slot)
if inDegree < 1:
_logger.warn("slot %s [%s] have no options" % (slot, speci))
missedSlots.append(slot)
_logger.info('Constructed flow graph')
return missedSlots
def get_prev_aws_date(speaker):
lastDate = None
if speakersSpecialization_.get(speaker, 'UNSPECIFIED') == 'UNSPECIFIED':
_logger.warning("Could not find specialization for %s" % speaker)
# Last entry is most recent
if speaker not in aws_.keys():
# We are here because this speaker has not given any AWS yet.
freshers_.add(speaker)
# If this user has PHD/POSTDOC, or INTPHD title. We create a dummy
# last date to bring her into potential speakers.
# First make sure, I have their date of joining. Otherwise I
# can't continue. For MSc By Research/INTPHD assign their first AWS
# after 18 months. For PHD and POSTDOC, it should be after 12 months.
if speakers_[speaker]['title'] == 'INTPHD':
# InPhd should get their first AWS after 15 months of
# joining.
_logger.info('%s = INTPHD with 0 AWS so far' % speaker)
joinDate = speakers_[speaker]['joined_on']
if not joinDate:
_logger.warn("Could not find joining date")
else:
lastDate = monthdelta(joinDate, +6)
if speakers_[speaker]['title'] == 'MSC':
# MSc should get their first AWS after 18 months of
# joining. Same as INTPHD
_logger.info('%s = MSC BY RESEARCH with 0 AWS so far' % speaker)
joinDate = speakers_[speaker]['joined_on']
if not joinDate:
_logger.warn("Could not find joining date")
else:
lastDate = monthdelta(joinDate, +6)
elif speakers_[speaker]['title'] in ['PHD', 'POSTDOC']:
joinDate = speakers_[speaker]['joined_on']
_logger.info('%s PHD/POSTDOC with 0 AWS so far' % speaker)
if not joinDate:
_logger.warn("Could not find joining date")
else:
try:
# if datetime.
lastDate = joinDate.date()
except Exception as e:
# Else its date
lastDate = joinDate
else:
# We are here because this speaker has given AWS before
# If this speaker is already on upcoming AWS list, ignore it.
if speaker in upcoming_aws_:
_logger.info('Speaker %s is already scheduled on %s' %
(speaker, upcoming_aws_[speaker]))
return None
# If this speakers is MSC by research and has given AWS before, she/he
# need not give another.
elif speakers_[speaker]['title'] == 'MSC':
_logger.info('%s is MSC and has given AWS in the past' % speaker)
return None
else:
lastDate = aws_[speaker][-1]['date']
return lastDate