rycerzes's picture
Workbench v3: 16 families, 30 tasks in submission
914310a verified
Raw History Blame Contribute Delete
4.51 kB
# SPDX-License-Identifier: BSD-3-Clause
"""Industrial and physical systems: sequence jobs on one machine.
Minimize total weighted tardiness. Higher difficulty adds sequence-dependent
setup times, a maintenance window and a cancelled job (both in memo.txt). Any
optimal order earns 1.0; anything else 0 (a near-miss tier would reward random orders).
"""
import itertools
from ..sandbox import Workspace
from ..tasks.base import csv_text
from .base import Episode, Family, file_tools, submit_action, submit_hint
RULES = """# Scheduling rules
- One machine processes one job at a time, without interruption, in the order you submit.
- A job cannot start before its release time. Times are minutes from shift start (t = 0).
- Before each job the machine needs the setup time from setups.csv for (previous job -> this job);
the first job uses the row for IDLE. Without setups.csv there is no setup time.
- Setup and processing form one block. If a block would overlap a maintenance window, the
block starts when the window ends.
- Tardiness of a job = max(0, completion time - due time). Cost = sum of weight * tardiness.
- Submit every active job exactly once. memo.txt, when present, overrides the other files.
"""
def schedule_cost(order, jobs, setups, window):
t, prev, cost = 0, "IDLE", 0
for j in order:
release, proc, due, weight = jobs[j]
setup = setups.get((prev, j), 0) if setups else 0
start = max(t, release)
if window and start < window[1] and start + setup + proc > window[0]:
start = window[1]
t = start + setup + proc
cost += weight * max(0, t - due)
prev = j
return cost
def make(difficulty, rng):
n = {1: 4, 2: 5, 3: 6, 4: 7}[difficulty]
names = [f"J{i}" for i in range(1, n + 2)]
jobs = {}
t = 0
for name in names:
proc = rng.randrange(10, 60, 5)
release = rng.choice([0, 0, rng.randrange(0, max(10, t), 5)])
due = release + proc + rng.randrange(0, 120, 5)
jobs[name] = (release, proc, due, rng.randint(1, 5))
t += proc
# Relabel at random so the creation order (J1 first) is not a good default answer.
labels = rng.sample(names, len(names))
jobs = {labels[i]: jobs[names[i]] for i in range(len(names))}
names = labels
cancelled = rng.choice(names) if difficulty == 4 else None
if difficulty < 4:
jobs.pop(names[-1])
active = [j for j in jobs if j != cancelled]
setups = None
if difficulty >= 2:
setups = {(a, b): rng.randrange(0, 25, 5) for a in ["IDLE", *jobs] for b in jobs if a != b}
window = None
notes = []
if difficulty >= 3:
start = rng.randrange(40, 140, 10)
window = (start, start + rng.randrange(20, 50, 10))
notes.append(f"- Planned maintenance: the machine is down from t = {window[0]} to t = {window[1]}.")
if cancelled:
notes.append(f"- {cancelled} was cancelled by the customer. Do not schedule it.")
costs = {order: schedule_cost(order, jobs, setups, window) for order in itertools.permutations(active)}
best = min(costs.values())
files = {
"rules.md": RULES,
"jobs.csv": csv_text(["job", "release_min", "processing_min", "due_min", "weight"], [[j, *jobs[j]] for j in jobs]),
}
if setups:
files["setups.csv"] = csv_text(["from", "to", "setup_min"], [[a, b, s] for (a, b), s in setups.items()])
if notes:
files["memo.txt"] = "# Shift memo\n" + "\n".join(notes) + "\n- Coffee machine on floor 2 is out of order.\n"
ws = Workspace(files)
def grade(answer):
order = answer.get("order") if isinstance(answer, dict) else None
if not isinstance(order, list) or sorted(map(str, order)) != sorted(active):
return 0.0, f"order must list each active job once ({len(active)} jobs)"
cost = costs[tuple(map(str, order))]
if cost == best:
return 1.0, f"optimal (cost {cost})"
return 0.0, f"cost {cost}; not optimal"
return Episode(
question="Choose the job order that minimizes total weighted tardiness, following rules.md.",
tools=file_tools(ws, files),
answer_hint=submit_hint({"order": ["J2", "J1", "..."]}),
grade=grade,
workspace=ws,
oracle=[submit_action({"order": list(min(costs, key=costs.get))})],
)
FAMILY = Family("machine_schedule", "industrial_physical_systems", "Single-machine sequencing with setups, maintenance and cancellations", make)