API Reference

Complete API documentation for the Structured Stochasticity framework. This documentation is auto-generated from Python docstrings.

Package Contents

• Noise Injection Strategies
  • Injection State
  • Base Class
  • Gaussian Noise
  • Uniform Noise
  • Annealed Noise
  • Layer-Selective Injection
  • Once-Per-Generation Injection
  • Factory Function
• PyTorch Hooks
  • Hook Configuration
  • Hidden State Hook
  • Noisy Inference Wrapper
• Benchmark Tasks
  • Data Classes
  • Base Task Class
  • Tower of Hanoi
  • Arithmetic Sequence
  • Logical Deduction
  • Task Factory
• Evaluation & Aggregation
  • Result Data Classes
  • Trajectory Aggregator
  • Evaluator
• Experiment Runner
  • Experiment Configuration
  • Experiment Class
  • Quick Experiment
  • CLI Entry Point

Quick Reference

The main components of this framework are:

• injection - Noise injection strategies (Gaussian, Uniform, Annealed, etc.)

• hooks - PyTorch forward hooks for modifying hidden states

• tasks - Benchmark tasks (Tower of Hanoi, Arithmetic, Logical Deduction)

• evaluation - Trajectory aggregation and evaluation metrics

• experiment - Main experiment runner and configuration