Port SyntheticAssemblyGenerator to solver/datagen/generator.py #5

New Issue

forbes · 2026-02-02T19:32:42Z

forbes commented

2026-02-02 19:32:42 +00:00

Summary

Move SyntheticAssemblyGenerator class from data/synthetic/pebble-game.py (L1047-1275) into solver/datagen/generator.py.

Class overview

Generates assembly graphs with known minimal constraint sets for training data. Uses the pebble game to incrementally build assemblies, tracking which constraints are independent at each step.

Methods to port

__init__(seed) — initializes numpy RNG
generate_chain_assembly(n_bodies, joint_type) — serial kinematic chain (always underconstrained)
generate_rigid_assembly(n_bodies) — spanning tree with fixed joints, then relaxes to weaker types while maintaining rigidity
generate_overconstrained_assembly(n_bodies, extra_joints) — rigid assembly + extra redundant joints
generate_training_batch(batch_size, n_bodies_range) — generates labeled training examples with per-joint independence/redundancy flags

Requirements

Import from solver.datagen.types (RigidBody, Joint, JointType)
Import from solver.datagen.analysis (analyze_assembly)
Full type annotations
No behavioral changes
__all__ = ["SyntheticAssemblyGenerator"]

Depends on

#1 (shared types)
#4 (analyze_assembly)

## Summary Move `SyntheticAssemblyGenerator` class from `data/synthetic/pebble-game.py` (L1047-1275) into `solver/datagen/generator.py`. ## Class overview Generates assembly graphs with known minimal constraint sets for training data. Uses the pebble game to incrementally build assemblies, tracking which constraints are independent at each step. ## Methods to port - `__init__(seed)` — initializes numpy RNG - `generate_chain_assembly(n_bodies, joint_type)` — serial kinematic chain (always underconstrained) - `generate_rigid_assembly(n_bodies)` — spanning tree with fixed joints, then relaxes to weaker types while maintaining rigidity - `generate_overconstrained_assembly(n_bodies, extra_joints)` — rigid assembly + extra redundant joints - `generate_training_batch(batch_size, n_bodies_range)` — generates labeled training examples with per-joint independence/redundancy flags ## Requirements - [ ] Import from `solver.datagen.types` (RigidBody, Joint, JointType) - [ ] Import from `solver.datagen.analysis` (analyze_assembly) - [ ] Full type annotations - [ ] No behavioral changes - [ ] `__all__ = ["SyntheticAssemblyGenerator"]` ## Depends on - #1 (shared types) - #4 (analyze_assembly)

forbes added the phase:1 port labels 2026-02-02 19:32:42 +00:00

forbes referenced this issue

2026-02-02 19:32:57 +00:00

Unit tests for ported datagen modules #6

forbes referenced this issue

2026-02-02 19:33:09 +00:00

Expand synthetic generator: parameterized assembly templates #7

forbes referenced this issue

2026-02-02 19:33:21 +00:00

Expand synthetic generator: geometric diversity #8

forbes referenced this issue

2026-02-02 19:33:53 +00:00