SMT Design Space Extension 🌳

Enhancing SMT with Hierarchical & Conditional Design Spaces

This package is an extension to the SMT (Surrogate Modeling Toolbox), offering a powerful framework to define and manage hierarchical, mixed-type, and conditionally active variables in design spaces.

---

🔍 What It Does

• Hierarchical variables: Support for nested conditional variables (e.g., a rotor configuration branch that only activates when use_rotor = yes).
• Mixed types: Handles continuous, integer or categorical variables uniformly.
• Conditional activation: Meta-variables powerfully control lower-level variable activation based on context.
• Graph-based design space representation: Clean and intuitive implementation of complex, branching designs.
• Extensible subtress: Easily add new types or layers of conditional logic.

---

📖 Why It Matters

In many engineering and simulation contexts, the design parameters form a structured, dependent hierarchy—optimizing them without a formal framework can lead to errors or inefficient models. This extension provides:

• A formal representation of hierarchical dependencies.
• Seamless integration with surrogate modeling tools and optimization routines.
• A foundation for improved experimental design and architectural exploration.

---

🔗 Reference

This implementation is based on:

Hierarchical Modeling and Architecture Optimization: Review and Unified Framework
P. Saves, E. Hallé‑Hannan, J. Bussemaker, Y. Diouane, N. Bartoli (June 2025). arXiv:2506.22621 (arXiv)

The paper provides a comprehensive survey and introduces a unified graph-based model that serves as the theoretical foundation for this extension.

---

📦 Installation

pip install smt-design-space-ext

https://pypi.org/project/smt-design-space-ext/

Requirements. See Requirements for details.

---

🚀 Quick Start Example

from smt_design_space_ext import (
    HAS_CONFIG_SPACE,
    HAS_ADSG,
    AdsgDesignSpaceImpl,
    ConfigSpaceDesignSpaceImpl,
    BaseDesignSpace,
    CategoricalVariable,
    FloatVariable,
    IntegerVariable,
    OrdinalVariable,
)

 ds = ConfigSpaceDesignSpaceImpl(
            [
                CategoricalVariable(["A", "B", "C"]),  # x0
                CategoricalVariable(["E", "F"]),  # x1
                IntegerVariable(0, 1),  # x2
                FloatVariable(0.1, 1),  # x3
            ],
            random_state=42,
        )
        ds.declare_decreed_var(
            decreed_var=3, meta_var=0, meta_value="A"
        )  # Activate x3 if x0 == A

This dynamically builds a tree-like structure of variables, enabling clear and constrained space exploration.

---

🧩 Integration

• Integrates readily with SMT’s Kriging modules.
• Compatible with Bayesian or gradient‑based optimizers.
• Prepare space definitions for use with surrogate modeling pipelines.

---

📚 Additional Resources

• The SMT repository contains tutorials and example notebooks to demonstrate usage.
• Issue tracker and usage discussions can be found under SMTorg/smt-design-space-ext.
• Author metadata and license summary confirm this is BSD‑licensed.

---

✅ Citation

If you use this package in research, please cite:

P. Saves, E. Hallé‑Hannan, J. Bussemaker, Y. Diouane, N. Bartoli,
“Hierarchical Modeling and Architecture Optimization: Review and Unified Framework,”
arXiv:2506.22621, June 2025.

---

📜 License

Distributed under the BSD-3-Clause license. See LICENSE for details.