tasq/node_modules/@claude-flow/neural/dist/domain/services/learning-service.js

/**
 * Learning Domain Service - Domain Layer
 *
 * Contains learning logic for pattern recognition and optimization.
 *
 * @module v3/neural/domain/services
 */
import { Pattern } from '../entities/pattern.js';
/**
 * Learning Domain Service
 */
export class LearningDomainService {
    patterns = new Map();
    /**
     * Extract patterns from trajectory
     */
    extractPatterns(trajectory) {
        const extracted = [];
        // Extract task-routing pattern
        if (trajectory.outcome === 'success') {
            const taskPattern = Pattern.create({
                type: 'task-routing',
                name: `route_${trajectory.id}`,
                description: `Learned from successful trajectory`,
                condition: this.extractCondition(trajectory.input),
                action: trajectory.actions[0] ?? 'default',
                confidence: 0.6 + trajectory.reward * 0.2,
                metadata: { source: trajectory.id },
            });
            extracted.push(taskPattern);
        }
        // Extract error recovery pattern if failure
        if (trajectory.outcome === 'failure' && trajectory.actions.length > 1) {
            const lastAction = trajectory.actions[trajectory.actions.length - 1];
            const recoveryPattern = Pattern.create({
                type: 'error-recovery',
                name: `recovery_${trajectory.id}`,
                description: `Recovery action from failure`,
                condition: `error:${trajectory.actions[0]}`,
                action: lastAction,
                confidence: 0.5,
                metadata: { source: trajectory.id },
            });
            extracted.push(recoveryPattern);
        }
        return extracted;
    }
    /**
     * Update patterns based on trajectory outcome
     */
    updatePatterns(trajectory) {
        let patternsUpdated = 0;
        let totalConfidenceChange = 0;
        for (const pattern of this.patterns.values()) {
            if (pattern.matches(trajectory.input)) {
                const oldConfidence = pattern.confidence;
                if (trajectory.outcome === 'success') {
                    pattern.recordSuccess();
                }
                else {
                    pattern.recordFailure();
                }
                totalConfidenceChange += pattern.confidence - oldConfidence;
                patternsUpdated++;
            }
        }
        // Extract and add new patterns
        const newPatterns = this.extractPatterns(trajectory);
        for (const pattern of newPatterns) {
            this.patterns.set(pattern.id, pattern);
        }
        return {
            patternsExtracted: newPatterns.length,
            patternsUpdated,
            confidenceChange: totalConfidenceChange,
        };
    }
    /**
     * Get route recommendation for task
     */
    getRouteRecommendation(taskDescription) {
        const matchingPatterns = [];
        for (const pattern of this.patterns.values()) {
            if (pattern.type !== 'task-routing')
                continue;
            if (pattern.matches(taskDescription)) {
                const score = pattern.confidence * (pattern.isReliable() ? 1.2 : 1.0);
                matchingPatterns.push({ pattern, score });
            }
        }
        // Sort by score
        matchingPatterns.sort((a, b) => b.score - a.score);
        if (matchingPatterns.length === 0) {
            return this.getDefaultRecommendation(taskDescription);
        }
        const best = matchingPatterns[0];
        const alternates = matchingPatterns.slice(1, 4).map((m) => ({
            role: m.pattern.action,
            confidence: m.score,
        }));
        return {
            agentRole: best.pattern.action,
            confidence: best.score,
            reasoning: `Based on pattern "${best.pattern.name}" with ${best.pattern.successCount} successes`,
            alternates,
        };
    }
    /**
     * Get default recommendation based on keywords
     */
    getDefaultRecommendation(task) {
        const taskLower = task.toLowerCase();
        const keywordMap = {
            code: 'coder',
            implement: 'coder',
            write: 'coder',
            test: 'tester',
            review: 'reviewer',
            plan: 'planner',
            research: 'researcher',
            security: 'security-architect',
            performance: 'performance-engineer',
            memory: 'memory-specialist',
        };
        for (const [keyword, role] of Object.entries(keywordMap)) {
            if (taskLower.includes(keyword)) {
                return {
                    agentRole: role,
                    confidence: 0.5,
                    reasoning: `Keyword match: "${keyword}"`,
                    alternates: [],
                };
            }
        }
        return {
            agentRole: 'coder',
            confidence: 0.3,
            reasoning: 'Default fallback',
            alternates: [],
        };
    }
    /**
     * Extract condition from input
     */
    extractCondition(input) {
        // Extract key terms from input
        const words = input.toLowerCase().split(/\s+/);
        const keyWords = words.filter((w) => w.length > 4).slice(0, 5);
        return keyWords.join('|');
    }
    /**
     * Consolidate patterns (merge duplicates, prune low-confidence)
     */
    consolidate(minConfidence = 0.3) {
        let merged = 0;
        let pruned = 0;
        const toRemove = [];
        for (const [id, pattern] of this.patterns) {
            // Prune low confidence patterns with enough data
            if (pattern.confidence < minConfidence && pattern.successCount + pattern.failureCount > 20) {
                toRemove.push(id);
                pruned++;
            }
        }
        for (const id of toRemove) {
            this.patterns.delete(id);
        }
        return { merged, pruned };
    }
    /**
     * Get all patterns
     */
    getPatterns() {
        return Array.from(this.patterns.values());
    }
    /**
     * Get patterns by type
     */
    getPatternsByType(type) {
        return Array.from(this.patterns.values()).filter((p) => p.type === type);
    }
    /**
     * Add pattern
     */
    addPattern(pattern) {
        this.patterns.set(pattern.id, pattern);
    }
    /**
     * Remove pattern
     */
    removePattern(id) {
        return this.patterns.delete(id);
    }
}
//# sourceMappingURL=learning-service.js.map