PME-toolkit (Python)

Python implementation of PME-toolkit for design-space dimensionality reduction in parametric shape optimization.

Supports:

• PME (Parametric Model Embedding)
• PI-PME (Physics-Informed PME)
• PD-PME (Physics-Driven PME)
• analytical backmapping to original design variables

---

Overview

The Python package provides a fully functional implementation of PME-based workflows using a JSON-driven interface for reproducible studies.

It is aligned with the MATLAB implementation and validated through cross-language regression tests.

---

Installation

From PyPI

pip install pme-toolkit

From source (development mode)

pip install -e python/

---

Command-line interface

Run a PME case:

pme-run tests/cases/test_glider.json

Run backmapping:

pme-back tests/cases/test_glider_back.json

The CLI:

• parses the JSON configuration
• executes the PME workflow
• writes outputs to the specified outdir
• ensures consistency with MATLAB results

---

Programmatic usage

Example:

from pme_toolkit.model import fit_from_case

model = fit_from_case("tests/cases/test_glider.json")

print(model.nconf)
print(model.alpha_train.shape)

---

Repository integration

The Python implementation is part of the full PME-toolkit repository:

• shared JSON configuration format
• shared datasets
• shared benchmark definitions
• validated against MATLAB through regression testing

---

Testing

From repository root:

pytest tests/python -q

Test suite covers:

• configuration loading
• PME workflows
• backmapping
• regression against MATLAB reference

---

Datasets

• lightweight test dataset:

  tests/data/

• benchmark datasets:

  databases/

---

Status

The Python implementation is fully functional for PME, PI-PME, and PD-PME within the current scope.

It is actively maintained and serves as a production-ready interface for reproducible PME workflows.