233 lines
8.3 KiB
Python
233 lines
8.3 KiB
Python
import json
|
|
from math import floor
|
|
import subprocess
|
|
|
|
|
|
|
|
# Maximum time for AK: 2 Timeslots
|
|
SLOT_LEN = 120
|
|
|
|
DAYS = 1
|
|
START = 480
|
|
END = 1080
|
|
|
|
|
|
class WorkGroup:
|
|
def __init__(self, id, interest, track, projector, reso, length, leaders, constraints):
|
|
self.id = id # must be int
|
|
self.interest = interest
|
|
self.track = track if track is not None else 'none' # haben alle aks einen track?
|
|
self.projector = 'true' if projector else 'false'
|
|
self.reso = 'true' if reso else 'false'
|
|
self.length = floor(length/SLOT_LEN) + 1
|
|
self.leaders = leaders
|
|
self.after_time = []
|
|
self.before_time = []
|
|
for constraint in constraints:
|
|
if constraint["type"] == 'OnlyAfterTime':
|
|
if constraint['workGroup'] is None:
|
|
self.after_time.append({'day': constraint['day'], 'time': constraint['time']})
|
|
elif constraint["type"] == 'OnlyBeforeTime':
|
|
self.before_time.append({'day': constraint['day'], 'time': constraint['time']})
|
|
|
|
def __str__(self):
|
|
leader_rules = ''
|
|
for leader in self.leaders:
|
|
leader_rules += f'leader({self.id}, "{leader}").\n'
|
|
return f'ak({self.id}, {self.interest}, "{self.track}", {self.projector}, {self.reso}, {self.length}).' + leader_rules
|
|
|
|
|
|
class Room:
|
|
def __init__(self, id, n_places, projector): # TODO internet, whiteboard & co.
|
|
self.id = id # must be int
|
|
self.n_places = n_places
|
|
self.projector = 'true' if projector else 'false'
|
|
|
|
def __str__(self):
|
|
return f'room({self.id}, {self.n_places}, {self.projector}).'
|
|
|
|
|
|
class Timeslot:
|
|
def __init__(self, order, timeslot):
|
|
self.begin, self.end, self.day = timeslot
|
|
self.id = f'{self.day}#{self.begin}#{self.end}'
|
|
self.order = order
|
|
# self.begin = begin
|
|
# self.end = end
|
|
# self.day = day
|
|
|
|
def __str__(self):
|
|
return f'timeslot("{self.id}", {self.order}, {self.begin}, {self.end}, {self.day}).'
|
|
|
|
class Schedule:
|
|
def __init__(self, id, workgroup, day, time, room):
|
|
self.id = id
|
|
self.workgroup = workgroup
|
|
self.room = room
|
|
self.day = day
|
|
self.time = time
|
|
|
|
def to_dict(self):
|
|
return {
|
|
'id': self.id,
|
|
'workGroup': self.workgroup,
|
|
'room': self.room,
|
|
'time': self.time,
|
|
'day': self.day,
|
|
'lockRoom': False,
|
|
'lockTime': False,
|
|
}
|
|
|
|
|
|
def parse_workgroups(data):
|
|
workgroups = []
|
|
id_to_workgroup = {}
|
|
for i, workgroup in enumerate(data['workGroups']):
|
|
workgroups.append(
|
|
WorkGroup(workgroup['id'], workgroup['interested'], workgroup['track'], workgroup['projector'],
|
|
workgroup['resolution'], workgroup['length'], workgroup['leader'], workgroup['constraints']))
|
|
id_to_workgroup[workgroup['id']] = i
|
|
return workgroups, id_to_workgroup
|
|
|
|
def parse_rooms(data):
|
|
rooms = []
|
|
id_to_room = {}
|
|
for i, room in enumerate(data['rooms']):
|
|
rooms.append(Room(room['id'], room['places'], room['projector'])) # TODO internet, whiteboard & co.
|
|
id_to_room[room['id']] = i
|
|
return rooms, id_to_room
|
|
|
|
|
|
#######################
|
|
# READ BACKUP JSON #
|
|
#######################
|
|
|
|
|
|
data = json.load(open('backup.json'))
|
|
|
|
workgroups, id_to_workgroup = parse_workgroups(data)
|
|
rooms, id_to_room = parse_rooms(data)
|
|
|
|
|
|
|
|
rules = """
|
|
%LENGTH = { schedule(AK,TIMESLOT,ROOM): timeslot(TIMESLOT,_,_,_,_), room(ROOM,_,_) } :- ak(AK,_,_,_,_,LENGTH).
|
|
LENGTH = { schedule(AK,TIMESLOT,ROOM): timeslot(TIMESLOT,_,_,_,_), room(ROOM,_,_) } :- ak(AK,_,_,_,_,LENGTH).
|
|
|
|
% forbid 2AKS in the same room at the same time
|
|
:- schedule(AK1,TIMESLOT,ROOM), schedule(AK2,TIMESLOT,ROOM), AK1 != AK2.
|
|
|
|
% forbid ak with beamer request in room without beamer
|
|
:- schedule(AK,_,ROOM), ak(AK,_,_,true,_,_), room(ROOM,_,false).
|
|
|
|
% forbid ak in room that is too small
|
|
:- schedule(AK,_,ROOM), ak(AK,INTEREST,_,_,_,_), room(ROOM,PLACES,_), PLACES<INTEREST.
|
|
|
|
% forbid two AKs with same leader at same timestep
|
|
:- schedule(AK1,TIMESTEP,_), schedule(AK2,TIMESTEP,_), leader(AK1,LEADER), leader(AK2,LEADER), AK1 != AK2.
|
|
|
|
% schedules of longer aks should be in the same room and in consecutive timeslots
|
|
schedule(AK,TIMESLOT2,ROOM) :- ak(AK,_,_,_,_,2), schedule(AK,TIMESLOT1,ROOM), timeslot(TIMESLOT1,ORDER1,_,_,DAY), timeslot(TIMESLOT2,ORDER2,_,_,DAY), timeslot(TIMESLOT3,ORDER3,_,_,_), ORDER2 != ORDER1+1, ORDER1 = ORDER3+1, not schedule(AK,TIMESLOT3,ROOM) .
|
|
|
|
"""
|
|
|
|
for workgroup in workgroups:
|
|
rules += str(workgroup) + '\n'
|
|
|
|
for room in rooms:
|
|
rules += str(room) + '\n'
|
|
|
|
|
|
timeslots = [(start, start+SLOT_LEN, day) for day in range(DAYS) for start in range(START, END, SLOT_LEN)]
|
|
|
|
for order, timeslot in enumerate(timeslots):
|
|
rules += str(Timeslot(order, timeslot)) + '\n'
|
|
|
|
def time_to_id(timeslots, day, time):
|
|
timeslots = list(filter(lambda ts: time>= ts[1][0] and time <= ts[1][1] and day == ts[1][2], enumerate(timeslots)))
|
|
return timeslots[0][0]
|
|
|
|
for workgroup in filter(lambda wg: wg.after_time, workgroups):
|
|
for after_time in workgroup.after_time:
|
|
timeslot_id = time_to_id(timeslots, after_time['day'], after_time['time'])
|
|
timeslot = timeslots[timeslot_id] # after this slot: ok
|
|
print(f':- schedule({workgroup.id}, TIMESLOT, _), ak({workgroup.id},_,_,_,_,_), timeslot(TIMESLOT, ORDER, _, _, _), ORDER < {timeslot_id + 1}.')
|
|
rules += f':- schedule({workgroup.id}, TIMESLOT, _), ak({workgroup.id},_,_,_,_,_), timeslot(TIMESLOT, ORDER, _, _, _), ORDER < {timeslot_id + 1}.\n'
|
|
|
|
rules += "#show schedule/3."
|
|
|
|
|
|
#######################
|
|
# GENERATE SCHEDULE #
|
|
#######################
|
|
|
|
output_file_name = "generated_rules.pl"
|
|
|
|
open(output_file_name, 'w').write(rules)
|
|
|
|
# create subprocess definition
|
|
process = ['clingo', output_file_name]
|
|
|
|
# run subprocess, not sure, if busy-waiting..
|
|
completed_process = subprocess.run(process, universal_newlines = True, stdout = subprocess.PIPE, stderr = subprocess.DEVNULL)
|
|
output = completed_process.stdout
|
|
|
|
#######################
|
|
# OUTPUT SCHEDULE #
|
|
#######################
|
|
|
|
if 'UNSATISFIABLE' in output:
|
|
print('UNSATISFIABLE')
|
|
print(output)
|
|
else:
|
|
print('SATISFIABLE, outputting schedule...')
|
|
import re
|
|
try:
|
|
match = re.match('(.|\n)*Answer: (.+)\n(.*)\nSATISFIABLE', output)
|
|
schedules_string = match.group(3)
|
|
except Exception as e:
|
|
print(e)
|
|
print(output)
|
|
schedules = []
|
|
from collections import defaultdict
|
|
workgroup_to_schedule = defaultdict(list)
|
|
for i, schedule in enumerate(schedules_string.split('schedule')[1:]):
|
|
match = re.match('\((.+),"(.+)#(.+)#.+",(.+)\)', schedule)
|
|
try:
|
|
workgroup_id = id_to_workgroup[int(match.group(1))]
|
|
room_id = id_to_room[int(match.group(4))]
|
|
schedules.append(Schedule(i, data['workGroups'][workgroup_id], match.group(2), match.group(3), data['rooms'][room_id]))
|
|
# {
|
|
# 'id': i,
|
|
# 'workGroup': data['workGroups'][workgroup_id],
|
|
# 'room': data['rooms'][room_id],
|
|
# 'time': match.group(3),
|
|
# 'day': match.group(2),
|
|
# 'lockRoom': False,
|
|
# 'lockTime': False,
|
|
# })
|
|
workgroup_to_schedule[workgroup_id].append(i)
|
|
except Exception as e:
|
|
print(e)
|
|
|
|
schedules_filtered = []
|
|
|
|
for workgroup_id in workgroup_to_schedule:
|
|
schedules_for_workgroup = workgroup_to_schedule[workgroup_id]
|
|
if len(schedules_for_workgroup) == 1:
|
|
schedules_filtered.append(schedules[schedules_for_workgroup[0]])
|
|
else:
|
|
# more than one schedule contains workgroup, because workgroup is 2 slots long
|
|
# find out which one is the first one and remove the second workgroup:
|
|
schedules_for_workgroup_0 = schedules[schedules_for_workgroup[0]]
|
|
schedules_for_workgroup_1 = schedules[schedules_for_workgroup[1]]
|
|
if schedules_for_workgroup_0.time < schedules_for_workgroup_1.time:
|
|
schedules_filtered.append(schedules_for_workgroup_0)
|
|
else:
|
|
schedules_filtered.append(schedules_for_workgroup_1)
|
|
|
|
|
|
data = {}
|
|
data['schedules'] = [s.to_dict() for s in schedules_filtered]
|
|
json.dump(data, open('output.json', 'w'))
|