attune-ai 2.0.0__py3-none-any.whl

attune_llm/learning/evaluator.py

@@ -0,0 +1,438 @@
+"""Session Evaluator for Continuous Learning
+
+Evaluates sessions to determine if they contain learnable patterns.
+Identifies high-value sessions worth analyzing for pattern extraction.
+
+Architectural patterns inspired by everything-claude-code by Affaan Mustafa.
+See: https://github.com/affaan-m/everything-claude-code (MIT License)
+See: ACKNOWLEDGMENTS.md for full attribution.
+
+Copyright 2025 Smart AI Memory, LLC
+Licensed under Fair Source 0.9
+"""
+
+from __future__ import annotations
+
+import logging
+import re
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import TYPE_CHECKING, Any
+
+if TYPE_CHECKING:
+    from attune_llm.state import CollaborationState
+
+logger = logging.getLogger(__name__)
+
+
+class SessionQuality(Enum):
+    """Quality rating for a session."""
+
+    EXCELLENT = "excellent"  # High learning value
+    GOOD = "good"  # Worth extracting patterns
+    AVERAGE = "average"  # Some value
+    POOR = "poor"  # Limited learning value
+    SKIP = "skip"  # Don't process
+
+
+@dataclass
+class SessionMetrics:
+    """Metrics computed for a session."""
+
+    interaction_count: int = 0
+    user_corrections: int = 0
+    successful_resolutions: int = 0
+    trust_delta: float = 0.0
+    empathy_level_avg: float = 0.0
+    error_mentions: int = 0
+    workaround_mentions: int = 0
+    preference_signals: int = 0
+    session_duration_minutes: float = 0.0
+
+    def to_dict(self) -> dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "interaction_count": self.interaction_count,
+            "user_corrections": self.user_corrections,
+            "successful_resolutions": self.successful_resolutions,
+            "trust_delta": self.trust_delta,
+            "empathy_level_avg": self.empathy_level_avg,
+            "error_mentions": self.error_mentions,
+            "workaround_mentions": self.workaround_mentions,
+            "preference_signals": self.preference_signals,
+            "session_duration_minutes": self.session_duration_minutes,
+        }
+
+
+@dataclass
+class EvaluationResult:
+    """Result of session evaluation."""
+
+    quality: SessionQuality
+    score: float  # 0.0 to 1.0
+    metrics: SessionMetrics
+    learnable_topics: list[str] = field(default_factory=list)
+    recommended_extraction: bool = False
+    reasoning: str = ""
+
+    def to_dict(self) -> dict[str, Any]:
+        """Convert to dictionary."""
+        return {
+            "quality": self.quality.value,
+            "score": self.score,
+            "metrics": self.metrics.to_dict(),
+            "learnable_topics": self.learnable_topics,
+            "recommended_extraction": self.recommended_extraction,
+            "reasoning": self.reasoning,
+        }
+
+
+class SessionEvaluator:
+    """Evaluates sessions for learning potential.
+
+    Analyzes collaboration sessions to determine if they contain
+    patterns worth extracting for continuous learning.
+    """
+
+    # Patterns indicating user corrections
+    CORRECTION_PATTERNS = [
+        r"actually[,\s]",
+        r"i meant",
+        r"no[,\s]+i",
+        r"that's not what i",
+        r"let me clarify",
+        r"to be clear",
+        r"what i really",
+        r"i should have said",
+        r"correction:",
+    ]
+
+    # Patterns indicating errors/debugging
+    ERROR_PATTERNS = [
+        r"error",
+        r"exception",
+        r"failed",
+        r"doesn't work",
+        r"not working",
+        r"broken",
+        r"bug",
+        r"issue",
+        r"problem",
+        r"crash",
+    ]
+
+    # Patterns indicating workarounds
+    WORKAROUND_PATTERNS = [
+        r"workaround",
+        r"instead[,\s]",
+        r"alternative",
+        r"hack",
+        r"trick",
+        r"bypass",
+        r"work around",
+        r"quick fix",
+    ]
+
+    # Patterns indicating preferences
+    PREFERENCE_PATTERNS = [
+        r"i prefer",
+        r"i like",
+        r"i don't like",
+        r"i always",
+        r"i never",
+        r"my style",
+        r"usually i",
+        r"please always",
+        r"please don't",
+    ]
+
+    # Patterns indicating successful resolution
+    SUCCESS_PATTERNS = [
+        r"that works",
+        r"perfect",
+        r"thanks",
+        r"great",
+        r"solved",
+        r"fixed",
+        r"working now",
+        r"exactly what i",
+    ]
+
+    def __init__(
+        self,
+        min_interactions: int = 3,
+        min_score_for_extraction: float = 0.4,
+    ):
+        """Initialize the evaluator.
+
+        Args:
+            min_interactions: Minimum interactions for evaluation
+            min_score_for_extraction: Minimum score to recommend extraction
+        """
+        self._min_interactions = min_interactions
+        self._min_score_for_extraction = min_score_for_extraction
+
+        # Compile patterns
+        self._correction_re = self._compile_patterns(self.CORRECTION_PATTERNS)
+        self._error_re = self._compile_patterns(self.ERROR_PATTERNS)
+        self._workaround_re = self._compile_patterns(self.WORKAROUND_PATTERNS)
+        self._preference_re = self._compile_patterns(self.PREFERENCE_PATTERNS)
+        self._success_re = self._compile_patterns(self.SUCCESS_PATTERNS)
+
+    def _compile_patterns(self, patterns: list[str]) -> re.Pattern:
+        """Compile patterns into a single regex."""
+        combined = "|".join(f"({p})" for p in patterns)
+        return re.compile(combined, re.IGNORECASE)
+
+    def evaluate(
+        self,
+        state: CollaborationState,
+    ) -> EvaluationResult:
+        """Evaluate a collaboration session.
+
+        Args:
+            state: The collaboration state to evaluate
+
+        Returns:
+            EvaluationResult with quality rating and metrics
+        """
+        # Compute metrics
+        metrics = self._compute_metrics(state)
+
+        # Determine learnable topics
+        topics = self._identify_topics(state, metrics)
+
+        # Calculate score
+        score = self._calculate_score(metrics, len(topics))
+
+        # Determine quality
+        quality = self._score_to_quality(score)
+
+        # Build reasoning
+        reasoning = self._build_reasoning(metrics, topics, score)
+
+        return EvaluationResult(
+            quality=quality,
+            score=score,
+            metrics=metrics,
+            learnable_topics=topics,
+            recommended_extraction=score >= self._min_score_for_extraction,
+            reasoning=reasoning,
+        )
+
+    def _compute_metrics(self, state: CollaborationState) -> SessionMetrics:
+        """Compute session metrics."""
+        metrics = SessionMetrics()
+
+        metrics.interaction_count = len(state.interactions)
+
+        if metrics.interaction_count == 0:
+            return metrics
+
+        # Calculate duration
+        if state.interactions:
+            first = state.interactions[0].timestamp
+            last = state.interactions[-1].timestamp
+            duration = (last - first).total_seconds() / 60.0
+            metrics.session_duration_minutes = duration
+
+        # Calculate trust delta
+        if state.trust_trajectory:
+            initial = state.trust_trajectory[0] if state.trust_trajectory else 0.5
+            final = state.trust_level
+            metrics.trust_delta = final - initial
+
+        # Calculate average empathy level
+        if state.level_history:
+            metrics.empathy_level_avg = sum(state.level_history) / len(state.level_history)
+        else:
+            metrics.empathy_level_avg = state.current_level
+
+        # Analyze user messages
+        for interaction in state.interactions:
+            if interaction.role != "user":
+                continue
+
+            content_lower = interaction.content.lower()
+
+            # Count corrections
+            if self._correction_re.search(content_lower):
+                metrics.user_corrections += 1
+
+            # Count error mentions
+            if self._error_re.search(content_lower):
+                metrics.error_mentions += 1
+
+            # Count workaround mentions
+            if self._workaround_re.search(content_lower):
+                metrics.workaround_mentions += 1
+
+            # Count preference signals
+            if self._preference_re.search(content_lower):
+                metrics.preference_signals += 1
+
+            # Count successful resolutions
+            if self._success_re.search(content_lower):
+                metrics.successful_resolutions += 1
+
+        return metrics
+
+    def _identify_topics(
+        self,
+        state: CollaborationState,
+        metrics: SessionMetrics,
+    ) -> list[str]:
+        """Identify learnable topics from the session."""
+        topics = []
+
+        if metrics.user_corrections > 0:
+            topics.append("user_corrections")
+
+        if metrics.error_mentions > 0 and metrics.successful_resolutions > 0:
+            topics.append("error_resolution")
+
+        if metrics.workaround_mentions > 0:
+            topics.append("workarounds")
+
+        if metrics.preference_signals > 0:
+            topics.append("preferences")
+
+        # Check for project-specific patterns
+        if self._has_project_specific_content(state):
+            topics.append("project_specific")
+
+        return topics
+
+    def _has_project_specific_content(self, state: CollaborationState) -> bool:
+        """Check if session contains project-specific content."""
+        # Look for file paths, function names, etc.
+        project_indicators = [
+            r"\.(py|js|ts|tsx|jsx|go|rs|java|cpp|c|h)\b",  # File extensions
+            r"def\s+\w+",  # Python functions
+            r"function\s+\w+",  # JS functions
+            r"class\s+\w+",  # Class definitions
+            r"import\s+",  # Imports
+            r"from\s+\w+\s+import",  # Python imports
+        ]
+
+        project_re = re.compile("|".join(project_indicators), re.IGNORECASE)
+
+        for interaction in state.interactions:
+            if project_re.search(interaction.content):
+                return True
+
+        return False
+
+    def _calculate_score(
+        self,
+        metrics: SessionMetrics,
+        topic_count: int,
+    ) -> float:
+        """Calculate overall learning score."""
+        score = 0.0
+
+        # Base score from interaction count
+        if metrics.interaction_count >= self._min_interactions:
+            score += 0.1
+
+        # Corrections are highly valuable
+        if metrics.user_corrections > 0:
+            score += min(metrics.user_corrections * 0.15, 0.3)
+
+        # Error resolutions are valuable
+        if metrics.error_mentions > 0 and metrics.successful_resolutions > 0:
+            resolution_rate = metrics.successful_resolutions / max(metrics.error_mentions, 1)
+            score += resolution_rate * 0.2
+
+        # Workarounds are valuable
+        if metrics.workaround_mentions > 0:
+            score += min(metrics.workaround_mentions * 0.1, 0.2)
+
+        # Preferences help personalization
+        if metrics.preference_signals > 0:
+            score += min(metrics.preference_signals * 0.05, 0.1)
+
+        # Topic diversity bonus
+        score += topic_count * 0.05
+
+        # Trust increase indicates good interactions
+        if metrics.trust_delta > 0:
+            score += min(metrics.trust_delta * 0.2, 0.1)
+
+        # Penalize very short sessions
+        if metrics.interaction_count < self._min_interactions:
+            score *= 0.5
+
+        return min(score, 1.0)
+
+    def _score_to_quality(self, score: float) -> SessionQuality:
+        """Convert score to quality rating."""
+        if score >= 0.7:
+            return SessionQuality.EXCELLENT
+        elif score >= 0.5:
+            return SessionQuality.GOOD
+        elif score >= 0.3:
+            return SessionQuality.AVERAGE
+        elif score >= 0.1:
+            return SessionQuality.POOR
+        else:
+            return SessionQuality.SKIP
+
+    def _build_reasoning(
+        self,
+        metrics: SessionMetrics,
+        topics: list[str],
+        score: float,
+    ) -> str:
+        """Build human-readable reasoning."""
+        parts = []
+
+        if metrics.user_corrections > 0:
+            parts.append(f"{metrics.user_corrections} user correction(s) found")
+
+        if metrics.successful_resolutions > 0:
+            parts.append(f"{metrics.successful_resolutions} successful resolution(s)")
+
+        if metrics.workaround_mentions > 0:
+            parts.append(f"{metrics.workaround_mentions} workaround(s) discussed")
+
+        if metrics.preference_signals > 0:
+            parts.append(f"{metrics.preference_signals} preference signal(s)")
+
+        if topics:
+            parts.append(f"Learnable topics: {', '.join(topics)}")
+
+        if not parts:
+            parts.append("Limited learning signals detected")
+
+        return "; ".join(parts) + f". Score: {score:.2f}"
+
+    def should_extract_patterns(
+        self,
+        state: CollaborationState,
+    ) -> bool:
+        """Quick check if patterns should be extracted.
+
+        Args:
+            state: Collaboration state to check
+
+        Returns:
+            True if extraction is recommended
+        """
+        result = self.evaluate(state)
+        return result.recommended_extraction
+
+    def get_extraction_priority(
+        self,
+        state: CollaborationState,
+    ) -> int:
+        """Get extraction priority (higher = more important).
+
+        Args:
+            state: Collaboration state to evaluate
+
+        Returns:
+            Priority from 0-100
+        """
+        result = self.evaluate(state)
+        return int(result.score * 100)