Expand description
Learning-to-Stop Models for WAND/HNSW Early Termination
Implements machine learning models for adaptive early stopping in WAND queries and HNSW vector search with confidence-based termination.
Target: Dynamic threshold learning with >90% accuracy per TODO.md
Structs§
- Calibration
Params - Confidence calibration parameters
- Confidence
Feature Extractor - Confidence
Model - Confidence-based termination model
- Confidence
Training Sample - Training sample for confidence model
- Feature
Extractors - Feature extractors for different components
- Hnsw
Candidate - HNSW candidate representation
- Hnsw
Feature Extractor - Hnsw
Learning Config - HNSW learning configuration
- Hnsw
Search State - Search state for HNSW queries
- Hnsw
Stopping Predictor - HNSW stopping predictor
- Hnsw
Training Sample - Training sample for HNSW predictor
- Learned
Stopping Decision - Stopping decision with learned confidence
- Learning
Config - Configuration for learning models
- Learning
Metrics - Learning metrics and performance tracking
- Learning
Stop Model - Learning-to-stop model coordinator
- Linear
Predictor - Linear predictor for confidence calibration
- Query
Context - Query context for feature extraction
- Stopping
Reasoning - Reasoning for stopping decisions
- Training
Scheduler - Training scheduler for model updates
- Wand
Feature Extractor - Feature extractors implementations
- Wand
Learning Config - WAND learning configuration
- Wand
Search State - Search state for WAND queries
- Wand
Stopping Predictor - WAND stopping predictor using learned features
- Wand
Training Sample - Training sample for WAND predictor
Enums§
- Confidence
Feature - Confidence feature types
- Hnsw
Feature - HNSW feature types for learning
- Wand
Feature - WAND feature types for learning