vykrum/hywe-training-data

HYWE Architectural Training Data

A structured, synthetic dataset of procedural architectural programming, design intent, and topological spatial configurations.

The dataset is generated through the HYWE Core Engine, a dependency-free computational core for discrete spatial representation and deterministic topological resolution. Design-intent narratives and HYWE Syntax representations are captured through the HYWE ecosystem and structured by the Hynteract data pipeline.

The dataset provides a research resource for investigating the relationship between linguistic descriptions of architectural intent, structured spatial programming, and deterministic topological configurations. It may also provide a foundation for future machine-learning research in architectural design and spatial reasoning.

• Ecosystem Repository: github.com/vykrum/Hywe
• Experimental Environment: hywe.in

---

Research Context

Rather than beginning with fixed geometry, HYWE explores how programmatic architectural intent may be expressed through hierarchy, adjacency, sequence, and relational structure.

HYWE treats spatial organization as a discrete computational problem in which relationships and programmatic constraints are resolved before geometric realization. Its spatial representation is based on a hybrid orthogonal-hexagonal grid (Hygrid) and integer-based spatial computation.

The dataset therefore does not primarily represent conventional architectural drawings or geometric models. It records structured relationships between:

design intent → HYWE Syntax → deterministic topology → spatial configuration

This makes the dataset useful for investigating whether architectural intent can be represented, learned, transformed, or evaluated through structured spatial representations rather than through geometry or imagery alone.

---

System Architecture Flow

Designer Intent
      ↓
HYWE Syntax
      ↓
Deterministic Topology
      ↓
Spatial Configuration
      ↓
Hynteract Structuring
      ↓
JSONL Dataset

The HYWE Core Engine remains deterministic and computationally separate from the probabilistic processes that may subsequently be applied to the dataset.

---

Data Curation & Computational Pipeline

Data is captured through a computational pipeline in which the deterministic spatial logic of HYWE is separated from the dataset-collection and structuring process.

• 1. Interactive Inputs: Node Tree Input & Boundary Editor
• 2. Core Representation: HYWE Syntax
• 3. Engine Resolution Pipeline:
  • Lexel: Architectural Programming and Flow Parsing
  • Hexel: Atomic Spatial Primitive
  • Coxel: Spatial Cluster Growth
  • Goxel: Geometry and Coordinate Processing
  • Xyxel: Planar Configuration
    • Branch: SVG Rendering & Spatial Analysis
    • Branch: Batch Processing (Feeds into Dataset)
  • Nexel: Spatial Nesting and Sub-configurations
  • Zaxel: Vertical Stacking and Volumetric Organization
    • Branch: WebGPU Massing
• 4. Dataset Structuring: The Batch Processing output is paired with the Design Intent Narrative to form the final Hynteract Dataset.

Data generation

1. Procedural Variation Designers construct or batch-generate spatial arrangements within HYWE.
2. Structured Representation Architectural intent and spatial constraints are represented using HYWE Syntax, the structured representation used by the HYWE computational system.
3. Deterministic Resolution The HYWE Core Engine resolves the representation into discrete spatial topology and configuration according to the selected sequence and spatial constraints.
4. Dataset Structuring Hynteract pairs the design-intent narrative with the corresponding HYWE Syntax and resolved configuration data.
5. Dataset Commitment Structured records are committed as JSON Lines (.jsonl) data to this Hugging Face repository.

---

The Computational Generation Pipeline

For a detailed explanation of the engine-specific terminology used below, including Lexel, Hexel, Coxel, Goxel, Xyxel, Nexel, and Zaxel, see the HYWE Wiki.

Every spatial configuration committed to this dataset is resolved using the HYWE Core Engine, whose spatial computation is implemented independently of external geometric optimization or physics engines.

The current computational pipeline comprises the following stages:

1. Lexel — Architectural Programming & Flow Parsing Role: Represents the structured architectural program, including hierarchy, relationships, sequence, and spatial constraints. Lexel provides the programmatic representation from which subsequent spatial resolution takes place.
2. Hexel — Atomic Spatial Primitive Role: Represents the fundamental discrete spatial unit within Hygrid. Hexel coordinates are represented using integer-based spatial computation, avoiding dependence on floating-point geometric coordinates for the underlying topological representation.
3. Coxel — Spatial Cluster Growth Role: Groups and evolves collections of Hexels into spatial clusters representing programmatic zones and relationships.
4. Goxel — Geometry & Coordinate Processing Role: Provides coordinate grids, geometric processing, and the computational mechanisms required to translate discrete spatial structures into geometric representations.
5. Xyxel — Planar Configuration Role: Resolves Coxel structures into planar spatial configurations and layouts.
6. Nexel — Spatial Nesting Role: Resolves nested spatial configurations and sub-configurations within parent spatial structures.
7. Zaxel — Vertical Organization Role: Extends planar configurations vertically, allowing spatial structures to be organized across multiple levels.

These stages form the computational path from structured architectural programming toward resolved spatial configurations. Not every dataset record necessarily exercises every stage to the same extent; the resulting configuration depends on the structure and scope of the source definition.

---

Data Schema

Each dataset record is stored in JSON Lines (.jsonl) format within the data/ directory. The top-level structure per record:

Field	Type	Description
definition	string	The single source of truth HYWE Syntax representation containing design constraints, block attributes, boundaries, and cell specs.
description	string	A natural language spatial narrative detailing the designer's intent, scale, flow, and typology.
configuration	string[]	An array of container-scoped geometry strings, each encoding 24 layout variations in a Base34-compressed format.

Complete Layout Sweeps: Although the definition field contains a default sequence rule in its Q attribute, representing the active configuration at the time of export, the dataset is not restricted to that single configuration. The accompanying configuration field contains the resolved layout strings for all 24 supported sequence permutations, preserving the deterministic configuration space associated with each definition.

Configuration Layout String Format

Each string in configuration encodes one architectural container (Level or Nest) along with all its 24 sweeps:

Marker(ID_1;ID_2;...;ID_N | Variation_0 | Variation_1 | ... | Variation_23)

• Marker: L0, L1, N1, N2, etc.
• IDs: Semicolon-separated list of prefix-stripped local IDs (e.g. 1;1.1;1.2).
• Variations: 24 blocks separated by |. Each block contains semicolon-separated node coordinates.
• Coordinates: Flat, comma-separated Base34 X,Y pairs per hexel (e.g. F,5,H,5 = cells at (15,5) and (17,5)).
• Positional Alignment: The coordinate groups within each variation block are strictly positionally aligned with the IDs list in the header. (e.g., the 3rd coordinate string in Variation 5 belongs to the 3rd node listed in the header).

Sequence Mappings (Indices 0 to 23)

The index of each layout string variation block corresponds to the following sequence rule:

Index	Sequence Name	Index	Sequence Name		Index	Sequence Name	Index	Sequence Name
0	VRCWEE	6	VRCWWW		12	HRCWNN	18	HRCWSS
1	VRCCEE	7	VRCCWW		13	HRCCNN	19	HRCCSS
2	VRCWSE	8	VRCWNW		14	HRCWNE	20	HRCWSW
3	VRCCSE	9	VRCCNW		15	HRCCNE	21	HRCCSW
4	VRCWSW	10	VRCWNE		16	HRCWSE	22	HRCWNW
5	VRCCSW	11	VRCCNE		17	HRCCSE	23	HRCCNW

---

Research Use

The dataset is intended to support research into structured representations of architectural intent and spatial organization.

Potential research directions include:

• linguistic descriptions of architectural programs and their structured representations;
• computational representation of spatial relationships;
• deterministic topological generation;
• design-space exploration;
• sequence-dependent spatial variation;
• machine learning for structured spatial reasoning;
• mappings between natural-language intent and computational spatial representations;
• evaluation of learned or generated representations against deterministic HYWE configurations.

The dataset should therefore be understood primarily as a structured computational representation of architectural design intent and spatial topology, rather than as a conventional collection of floor-plan images or geometric models.

---

Dataset Status

The dataset is an active research dataset and is expected to evolve as the HYWE system and its data-collection methods develop.

The current dataset size should not be interpreted as a mature machine-learning training corpus. Its present value lies in establishing and testing the structure, semantics, and provenance of the representation.

Future releases may expand the number and diversity of records, introduce additional architectural typologies, and provide more explicit provenance and validation metadata.

---

Provenance & Reproducibility

Dataset records are generated from the HYWE computational ecosystem.

Because HYWE is an actively evolving research project, future dataset releases should be interpreted in relation to the corresponding HYWE engine and dataset schema versions.

---

Citation

If you use this dataset or the HYWE engine in your research or projects, please cite:

@misc{hywe2026dataset,
  author = {Subbaiah, Vikram},
  title = {HYWE Architectural Training Data: A Structured Dataset of Procedural Architectural Programming and Topological Layouts},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/vykrum/hywe-training-data}}
}

---

License

To align with the open-science principles of the project while distinguishing the dataset from the underlying software, the HYWE ecosystem uses separate licensing terms.

Dataset This Dataset Card and the datasets contained in this Hugging Face repository are licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. You are free to share and adapt the dataset provided appropriate attribution is given.

HYWE Software The HYWE source code is licensed separately under the MIT License. The dataset license does not grant rights to the HYWE source code beyond those provided by its applicable software license.