Generators

Two layers: a single job (one DAG of operations) and a full IPPS instance (several jobs combined, each with its own machine assignments).

Job generation

from ipps_drl import JobGenerator, JobConfig

cfg = JobConfig(
    machine_range=(5, 5),      # min, max machines available
    mas_p=0.5,                  # P(a given machine can process a given op)
    randomize_opes=False,       # if True, draw ope_num from `ope_range`
    ope_num=10,                 # base operation count
    or_num=2,                   # number of OR splits to inject
    and_p=0.3,                  # P(an OR branch gets nested AND branches)
    time_range=(100, 500),      # per-op mean processing time
)
job = JobGenerator(cfg).generate()
print(job.num_machines, job.num_opes)
print(job.lines[:4])

# Save with the canonical "_mas_<N>.txt" suffix the pool loader expects:
job.save("data/jobs/job_with_mas_5/job_42")  # → "job_42_mas_5.txt"

JobConfig knobs

group	fields
Machines	machine_range, mas_p, time_range, time_bias
DAG skeleton	max_out, alpha, beta
Operation count	randomize_opes, ope_num, ope_range, total_ope_range
OR splits	or_num, or_p, or_road_num, ope_num_orpath
AND inside OR	and_p, and_road_num, ope_num_andpath

Instance generation

from ipps_drl import InstanceGenerator, InstanceConfig

gen = InstanceGenerator(InstanceConfig(num_jobs=6, num_machines=5))

# Sample 6 jobs from a pre-generated pool on disk
inst = gen.from_pool("data/jobs/job_with_mas_5/")

# Or roll fresh jobs and combine on the fly
inst = gen.from_fresh_jobs(JobConfig(or_num=2))

inst.save("my_problem.ipps")
print(inst.num_jobs, inst.num_machines, inst.num_opes)
print(inst.source_job_ids)        # populated by from_pool

from_pool re-rolls until the selected jobs collectively use exactly num_machines machines. The cap is controlled by InstanceConfig.max_resample_attempts (None = unbounded).

Legacy CaseGenerator

For backwards compatibility with scripts/train_drl.py and any of your own old scripts:

from ipps_drl import CaseGenerator
case = CaseGenerator(job_num=5, machine_num=3, job_folder="data/jobs/job_with_mas_3/")
lines = case.get_case()           # list[str], same format as before

Internally this just wraps InstanceGenerator.

Bundled CLIs

# Roll a fresh job pool (4500 jobs at 5 machines each)
python -m ipps_drl.generator.jobs_generator \
    --out_dir data/jobs/job_with_mas_5 --count 4500 --machines 5 --max_or 3

# Assemble 100 IPPS instances from a pool
python -m ipps_drl.generator.case_generator_ipps \
    --job_folder data/jobs/job_with_mas_5 --num_jobs 6 --num_machines 5 --count 100

Both honour --seed.