doin-plugins

Domain plugins for DOIN (Decentralized Optimization and Inference Network).

Included Plugins

Quadratic (Reference)

Simple quadratic function optimization — no ML frameworks needed. Used for testing the full DOIN pipeline.

• Optimizer: Hill-climbing on f(x) = Σ(x_i - target_i)²
• Inferencer: Evaluates parameters against target
• Synthetic Data: Generates noisy target variants for verification

Predictor (ML — DEAP GA Wrapper)

Wraps harveybc/predictor timeseries prediction system. Requires TensorFlow.

The predictor plugin wraps predictor's full DEAP genetic algorithm — it doesn't replace it. DOIN hooks into the GA via callbacks, adding decentralized island-model migration on top of the existing optimizer.

DEAP GA callback hooks:

• on_generation_start — called before each generation (inject migrated champions here)
• on_generation_end — called after each generation (collect stats, check convergence)
• on_between_candidates — called between candidate evaluations within a generation
• on_champion_found — called when a new best individual is found (broadcast to network)

5-stage incremental optimization: Predictor's GA runs 5 stages of increasing complexity. DOIN wraps all stages, broadcasting champions at each stage boundary. Other nodes can inject received champions into their populations at any stage.

Champion broadcasting and migration injection: When on_champion_found fires, DOIN broadcasts the champion's parameters as an optimae. Other nodes receiving this optimae inject the champion into their local DEAP population, creating new genetic material for crossover — the classic island model.

Three-Level Patience System:

Level	Config Key	Controls	Default
L1 — Candidate Training	early_patience	Keras early stopping per candidate	80–100
L2 — Stage Progression	optimization_patience	GA generations before advancing stage	8–10
L3 — Meta-Optimizer	(not yet implemented)	Network-level param→performance predictor	—

• Optimizer: Wraps predictor's DEAP GA with callback hooks for island-model migration
• Inferencer: Evaluates model on test/synthetic data
• Synthetic Data: Wraps harveybc/timeseries-gan (SC-VAE-GAN) with block bootstrap fallback

Entry Points

doin.optimization/quadratic → QuadraticOptimizer
doin.optimization/predictor → PredictorOptimizer
doin.inference/quadratic → QuadraticInferencer
doin.inference/predictor → PredictorInferencer
doin.synthetic_data/quadratic → QuadraticSyntheticData
doin.synthetic_data/predictor → PredictorSyntheticData

Install

pip install git+https://github.com/harveybc/doin-core.git
pip install git+https://github.com/harveybc/doin-plugins.git

Tests

python -m pytest tests/ -v
# 33 tests passing (7 e2e lifecycle + unit + integration)

Key Tests

• test_e2e_lifecycle.py — Full optimae lifecycle: optimize → commit → reveal → quorum → verify → incentive → reputation (7 tests)
• test_plugins.py — Quadratic plugin unit tests
• test_network_integration.py — Multi-component integration

Part of DOIN

• doin-core — Consensus, models, crypto
• doin-node — Unified node
• doin-plugins — This package