DispatchAtlas¶
DispatchAtlas is a Python library for reproducible benchmarking of scheduling algorithms. One data model covers machine scheduling, resource-constrained project scheduling, and the placement of tasks across an edge-fog-cloud continuum. Benchmark families generate problems from a seed, and published suites such as Taillard's are rebuilt exactly. Every solver, from dispatching rules to metaheuristics and exact models, runs under the same stop criteria and counts its evaluations the same way, so budgets are comparable. Campaigns run every solver on every problem with paired seeds, record each run, resume after an interruption, and replay runs to check that they reproduce. The analysis turns a campaign into tables and figures, with every p-value next to an effect size and a confidence interval.
from dispatchatlas.benchmarks import generate
from dispatchatlas.core import check_schedule
from dispatchatlas.solvers import solve
problem = generate("job-shop", seed=1, jobs=6, machines=4)
result = solve(problem, "tabu-search", max_evaluations=2000, seed=3)
print(result.status, result.value, result.evaluations, result.stop_reason)
print(check_schedule(problem, result.schedule))
What the library provides¶
- Data model: tasks with alternative modes, resources of any capacity, dependencies with time lags and network transfers, release times, due dates, hard deadlines, setup times, and permutation schedules, with precise semantics and a feasibility checker that is independent of the solvers.
- Benchmarks: seeded benchmark families, published suites rebuilt offline or read from their files, instance catalogs written to disk with checksums, and descriptive instance features.
- Solvers: heuristics, local search, population and swarm methods, multi-objective algorithms, learning methods, and exact solvers behind one contract with exact budgets and seeded determinism.
- Campaigns: experiments described in a file, with fair comparisons verified from the plan, paired seeds, hard time limits, resumption, and replay.
- Analysis: rank tests, corrections, effect sizes, bootstrap intervals, quality indicators, tables, and figures, collected in a reproducible bundle.
Where to go next¶
- Getting started installs the library and walks through a first problem and a first campaign.
- The solver catalogue and the benchmark catalogue list everything that is registered, generated from the code.
- The command line reference documents every command.
- Extending shows how to add a solver, a benchmark family, or an objective, in a separate package if you like.
- The API reference documents every public module.