Data Pipeline & Loaders

What problem does it solve?

This skill provides a complete, phase-aware data-loading infrastructure for the brain_ai training pipeline, enabling consistent loaders, synthetic dev data, and centralized dataset management across all seven training phases.

Core Features & Use Cases

• Implements 7-phase loaders (SNN, Encoders, HTM, Workspace, Active Inference, Reasoning, Meta-Learning) with deterministic splits and synthetic data for development.
• Offers a BasePhaseLoader contract, sample-shape inference utilities, and ready-made collate functions to unify multimodal batches.
• Provides a DatasetRegistry to register, instantiate, list, download, and cache datasets across phases, with production fallbacks to synthetic data when needed.

Quick Start

Instantiate a phase loader (e.g., SNNPhaseLoader) with a dev DataConfig and start training by iterating its train loader.