design-research-problems 0.2.0

A compendium of canonical design research problems

design-research-problems

design-research-problems is a compact library and compendium of design research problems. It packages canonical research prompts, optimization benchmarks, and discrete grammar-style problems behind a small, typed Python API.

Overview

The initial release centers on five problem families plus a linked ideation metadata catalog:

• Text problems for human-subjects studies and prompt packets
• Decision problems with typed discrete or empirical evaluation interfaces
• Optimization problems with typed bounds and representative built-in baselines
• Grammar problems that describe discrete design actions and optional evaluation adapters
• MCP problems that ingest upstream MCP tool servers for agent-style workflows
• An ideation catalog with prompt records, variants, families, and study summaries

The public problem model now centers on a shared Problem documentation base and a ComputableProblem evaluation layer, with family-specific subclasses on top. Problem-family interchangeability is defined by those family base classes, not by matching catalog capability flags across different kinds.

The first catalog includes:

• 40 ideation-focused text prompts
• battery_pack_18650_open_ended_capacity_max for explicit 18650 layout-and-wiring capacity maximization under the shared battery backend
• gmpb_default_dynamic_min for a stateful dynamic minimization wrapper around the Generalized Moving Peaks Benchmark
• pill_capsule_min_area for constrained continuous optimization
• battery_pack_18650_series_parallel_cost_min for fixed-topology rectangular 18650 pack sizing under the shared battery backend
• planar_truss_span_mass_min, planar_truss_span_deflection_min, and planar_truss_span_fos_max for fixed-joint planar truss structural optimization under real trussme evaluation
• space_truss_span_mass_min for fixed-joint 3D space-truss structural optimization under the same shared truss backend
• planar_truss_span for the original seed planar truss grammar backed by trussme
• space_truss_span for a bounded 3D space-truss grammar backed by trussme
• battery_pack_18650_open_ended for explicit 18650 cell-by-cell graph-netlist co-design backed by an optional PyBaMM-shaped single-cell surrogate plus a library-owned pack solver
• battery_pack_18650_series_parallel for explicit 18650 series-parallel pack co-design backed by the same optional PyBaMM-shaped single-cell surrogate plus a library-owned pack solver
• six planar_roof_truss_* variants that approximate the roof-truss formulations reported by Shea and Cagan
• moneymaker_hip_pump_cost_min for a citation-backed scalarized pump benchmark
• treadle_pump_ide_material_min for a citation-backed scalarized treadle-pump benchmark
• mcp_build123d_parametric_mounting_bracket for MCP-ingested CAD workflows where agents author and evaluate Build123d scripts through a package-owned backend

Quickstart

Requires Python 3.12+. Install from PyPI with:

pip install design-research-problems

Install the optional trussme truss support with:

pip install "design-research-problems[grammar]"

The representative optimization baselines are included in the base install.

Install the optional external optimization backends with:

pip install "design-research-problems[solvers]"

The smooth continuous benchmarks continue to use SciPy baselines, while the open-ended battery co-design optimizer will automatically prefer pymoo, then nevergrad, and finally fall back to the built-in deterministic local search. The GMPB wrapper instead uses a simple random-search baseline because each evaluation advances a dynamic benchmark state.

Then inspect the catalog directly from the installed package:

python3 -c "from design_research_problems import list_problems; print(list_problems())"

And inspect the ideation corpus:

python3 -c "from design_research_problems import get_ideation_catalog; print(len(get_ideation_catalog().list_prompts()))"

Launch the packaged desktop GUIs with:

python3 -m design_research_problems.gui --app iot
python3 -m design_research_problems.gui --app truss

The IoT GUI renders a continuous room-temperature colorbar, and the truss GUI only evaluates structurally when the design is not under-determined.

For local development, reproducible installs are pinned to Python 3.12.12 in .python-version:

python3 -m venv .venv
source .venv/bin/activate
make dev
make ci

Docs

See the published documentation for the guide and reference layout. Build them locally with:

make docs

Contributing

Contribution guidelines live in CONTRIBUTING.md.