spt-reasoning-learning

Adaptive self-learning algorithms with pattern optimization

Overview

Implements algorithms for learning from task execution history, optimizing strategy selection, and improving agent performance over time. Sits between the core reasoning engine and storage layer, using stored patterns to derive actionable insights and compression-level recommendations.

Key Features

• Adaptive Learning -- AdaptiveLearner refines strategies from historical outcomes
• Async Learning -- AsyncLearnerV2 for non-blocking learning in concurrent workloads
• Compression Analysis -- CompressionAnalyzer derives optimal compression strategies per model
• Strategy Optimization -- Rank and select strategies with StrategyOptimizer
• Benchmarked -- Includes criterion benchmarks for learning performance

Dependencies

Key internal dependencies:

• spt-reasoning-core -- Pattern types, similarity scoring, compression trajectories
• spt-reasoning-storage -- Persistent storage for patterns and trajectories

Usage

use spt_reasoning_learning::{
    AdaptiveLearner, LearningConfig,
    CompressionAnalyzer, CompressionAnalyzerConfig,
    StrategyOptimizer,
};

let config = LearningConfig::default();
let learner = AdaptiveLearner::new(config);

API Reference

Core Types

Type	Description
AdaptiveLearner	Main learner that refines strategies from task history
LearningConfig	Configuration for adaptive learning
LearningInsight	A derived insight from pattern analysis
LearningStats	Aggregated learning statistics
AppliedLearning	Record of a learning application
AsyncLearnerV2	Non-blocking async learner (replaces deprecated AsyncLearner)
CompressionAnalyzer	Analyzes compression trajectories for optimal levels
CompressionAnalyzerConfig	Configuration for the compression analyzer
LevelAdjustment	Recommendation for compression level changes
StrategyOptimizer	Ranks and selects optimal strategies
StrategyRanking	Ranked list of strategies
OptimizationResult	Result of a strategy optimization run
LearningError	Error enum for learning operations