agentdb 1.0.12 → 1.1.0

package/dist/plugins/implementations/multi-task-learning.mjs

@@ -1,364 +0,0 @@
-/**
- * Multi-Task Learning Plugin
- *
- * Implements learning across multiple related tasks simultaneously,
- * leveraging shared representations and transfer learning.
- *
- * Key features:
- * - Shared and task-specific layers
- * - Task weighting and balancing
- * - Hard and soft parameter sharing
- * - Auxiliary task learning
- * - Task relationship modeling
- */
-import { BasePlugin } from '../base-plugin.mjs';
-/**
- * Multi-Task Learning Plugin Implementation
- */
-export class MultiTaskLearningPlugin extends BasePlugin {
-    constructor(config) {
-        super();
-        this.name = 'multi-task-learning';
-        this.version = '1.0.0';
-        this.experiences = [];
-        this.tasks = new Map();
-        this.sharingStrategy = 'hard_sharing';
-        // Shared representations
-        this.sharedWeights = new Map();
-        this.taskSpecificWeights = new Map();
-        // Task relationships
-        this.taskSimilarity = new Map();
-        this.auxiliaryTasks = new Set();
-        // Training parameters
-        this.uncertaintyWeighting = true;
-        this.gradientNormalization = true;
-        if (config) {
-            this.sharingStrategy = config.sharingStrategy || 'hard_sharing';
-            this.uncertaintyWeighting = config.uncertaintyWeighting !== false;
-        }
-        this.initializeSharedLayers();
-        // Mark as initialized for in-memory operation
-        this.initialized = true;
-    }
-    /**
-     * Override to skip initialization check for in-memory operation
-     */
-    checkInitialized() {
-        // No-op for multi-task learning - operates in-memory
-    }
-    /**
-     * Encode state to fixed dimension
-     */
-    encodeState(state) {
-        // Simple state encoding - ensure fixed dimension
-        const targetDim = 128;
-        if (state.length >= targetDim) {
-            return state.slice(0, targetDim);
-        }
-        // Pad if too short
-        const encoded = [...state];
-        while (encoded.length < targetDim) {
-            encoded.push(0);
-        }
-        return encoded;
-    }
-    /**
-     * Override selectAction to provide base implementation
-     */
-    async selectAction(state, context) {
-        // Simple default action selection
-        const stateArray = Array.isArray(state) ? state : [state];
-        const stateSum = stateArray.reduce((a, b) => a + b, 0);
-        const value = Math.tanh(stateSum / stateArray.length);
-        return {
-            id: String(Math.floor(Math.abs(value) * 10)),
-            embedding: stateArray,
-            confidence: Math.abs(value),
-        };
-    }
-    /**
-     * Override to store experiences in-memory without vectorDB
-     */
-    async storeExperience(experience) {
-        this.experiences.push(experience);
-        const taskId = experience.metadata?.taskId;
-        if (taskId && this.tasks.has(taskId)) {
-            // Update task performance
-            const task = this.tasks.get(taskId);
-            const alpha = 0.1;
-            const success = experience.reward > 0 ? 1 : 0;
-            task.performance = (1 - alpha) * task.performance + alpha * success;
-        }
-    }
-    /**
-     * Override to retrieve from local experiences
-     */
-    async retrieveSimilar(state, k) {
-        return this.experiences.slice(0, k).map((exp, idx) => ({
-            id: exp.id || `exp-${idx}`,
-            embedding: exp.state,
-            metadata: exp,
-            score: 1.0 - (idx * 0.1),
-        }));
-    }
-    /**
-     * Initialize shared layers
-     */
-    initializeSharedLayers() {
-        // Shared representation layers
-        for (let layer = 0; layer < 3; layer++) {
-            const size = [256, 128, 64][layer];
-            const weights = new Array(size).fill(0).map(() => Math.random() * 0.1 - 0.05);
-            this.sharedWeights.set(`shared_${layer}`, weights);
-        }
-    }
-    /**
-     * Add task to multi-task learning
-     */
-    addTask(taskId, name, priority = 1.0, isAuxiliary = false) {
-        this.tasks.set(taskId, {
-            id: taskId,
-            name,
-            priority,
-            weight: 1.0,
-            performance: 0,
-        });
-        if (isAuxiliary) {
-            this.auxiliaryTasks.add(taskId);
-        }
-        // Initialize task-specific layers
-        const taskLayers = new Map();
-        taskLayers.set('task_head', new Array(64).fill(0).map(() => Math.random() * 0.1 - 0.05));
-        this.taskSpecificWeights.set(taskId, taskLayers);
-    }
-    /**
-     * Compute task similarity
-     */
-    computeTaskSimilarity(task1Id, task2Id) {
-        const experiences1 = this.getTaskExperiences(task1Id);
-        const experiences2 = this.getTaskExperiences(task2Id);
-        if (experiences1.length === 0 || experiences2.length === 0) {
-            return 0.5;
-        }
-        // Compute average state embeddings
-        const avg1 = this.averageEmbedding(experiences1.map(e => e.state));
-        const avg2 = this.averageEmbedding(experiences2.map(e => e.state));
-        // Cosine similarity
-        return this.cosineSimilarity(avg1, avg2);
-    }
-    /**
-     * Get experiences for specific task
-     */
-    getTaskExperiences(taskId) {
-        // Simplified: return empty array
-        return [];
-    }
-    /**
-     * Average embedding
-     */
-    averageEmbedding(embeddings) {
-        if (embeddings.length === 0) {
-            return [];
-        }
-        const avg = new Array(embeddings[0].length).fill(0);
-        embeddings.forEach(emb => {
-            for (let i = 0; i < emb.length; i++) {
-                avg[i] += emb[i];
-            }
-        });
-        return avg.map(v => v / embeddings.length);
-    }
-    /**
-     * Cosine similarity
-     */
-    cosineSimilarity(a, b) {
-        let dot = 0;
-        let magA = 0;
-        let magB = 0;
-        for (let i = 0; i < Math.min(a.length, b.length); i++) {
-            dot += a[i] * b[i];
-            magA += a[i] * a[i];
-            magB += b[i] * b[i];
-        }
-        return dot / (Math.sqrt(magA) * Math.sqrt(magB) + 1e-8);
-    }
-    /**
-     * Compute automatic task weights using uncertainty
-     */
-    computeUncertaintyWeights() {
-        // Multi-task uncertainty weighting (Kendall et al., 2018)
-        const logVars = new Map();
-        this.tasks.forEach((task, taskId) => {
-            // Estimate log variance from task performance
-            const logVar = Math.log(1 - task.performance + 1e-8);
-            logVars.set(taskId, logVar);
-        });
-        // Compute weights
-        this.tasks.forEach((task, taskId) => {
-            const logVar = logVars.get(taskId);
-            task.weight = Math.exp(-logVar);
-        });
-    }
-    /**
-     * Normalize gradients across tasks
-     */
-    normalizeGradients(gradients) {
-        if (!this.gradientNormalization) {
-            return gradients;
-        }
-        const normalized = new Map();
-        // Compute gradient norms
-        const norms = new Map();
-        gradients.forEach((grad, taskId) => {
-            const norm = Math.sqrt(grad.reduce((sum, g) => sum + g * g, 0));
-            norms.set(taskId, norm);
-        });
-        // Normalize to unit norm
-        gradients.forEach((grad, taskId) => {
-            const norm = norms.get(taskId);
-            normalized.set(taskId, grad.map(g => g / (norm + 1e-8)));
-        });
-        return normalized;
-    }
-    /**
-     * Forward pass through shared and task-specific layers
-     */
-    async forwardMultiTask(state, taskId) {
-        let activation = this.encodeState(state);
-        // Shared layers
-        this.sharedWeights.forEach(weights => {
-            activation = this.forward(activation, weights);
-        });
-        // Task-specific layers
-        const taskWeights = this.taskSpecificWeights.get(taskId);
-        if (taskWeights) {
-            taskWeights.forEach(weights => {
-                activation = this.forward(activation, weights);
-            });
-        }
-        return activation[0] || 0;
-    }
-    /**
-     * Forward pass through layer
-     */
-    forward(input, weights) {
-        const output = new Array(weights.length).fill(0);
-        for (let i = 0; i < output.length; i++) {
-            output[i] = input[i % input.length] * weights[i];
-            output[i] = Math.tanh(output[i]); // Activation
-        }
-        return output;
-    }
-    /**
-     * Train across multiple tasks
-     */
-    async train(options) {
-        const startTime = Date.now();
-        const epochs = options?.epochs || 50;
-        const batchSize = options?.batchSize || 32;
-        let totalLoss = 0;
-        const taskLosses = new Map();
-        let experiencesProcessed = 0;
-        for (let epoch = 0; epoch < epochs; epoch++) {
-            // Update task weights if using uncertainty weighting
-            if (this.uncertaintyWeighting) {
-                this.computeUncertaintyWeights();
-            }
-            const taskGradients = new Map();
-            // Train on each task
-            for (const [taskId, task] of this.tasks) {
-                // Skip auxiliary tasks with low priority
-                if (this.auxiliaryTasks.has(taskId) && task.priority < 0.5) {
-                    continue;
-                }
-                // Retrieve task-specific experiences from local storage
-                const experiences = await this.retrieveSimilar(new Array(128).fill(0), batchSize);
-                const taskExps = experiences
-                    .filter(e => e.metadata?.metadata?.taskId === taskId)
-                    .map(e => ({
-                    state: e.metadata.state,
-                    action: { id: '0', embedding: e.metadata.state },
-                    reward: e.metadata?.reward || 0,
-                    nextState: e.metadata?.nextState || e.metadata.state,
-                    done: e.metadata?.done || false,
-                    metadata: e.metadata,
-                }));
-                let taskLoss = 0;
-                // Compute gradients for task
-                const gradients = new Array(256).fill(0);
-                for (const exp of taskExps) {
-                    const prediction = await this.forwardMultiTask(exp.state, taskId);
-                    const loss = Math.pow(prediction - exp.reward, 2);
-                    taskLoss += loss * task.weight;
-                    experiencesProcessed++;
-                    // Simplified gradient
-                    for (let i = 0; i < gradients.length; i++) {
-                        gradients[i] += (prediction - exp.reward) * exp.state[i % exp.state.length];
-                    }
-                }
-                taskGradients.set(taskId, gradients);
-                taskLosses.set(taskId, taskLoss / Math.max(1, taskExps.length));
-                totalLoss += taskLoss;
-            }
-            // Normalize gradients
-            const normalizedGradients = this.normalizeGradients(taskGradients);
-            // Update shared weights with aggregated gradients
-            normalizedGradients.forEach((gradients, taskId) => {
-                const task = this.tasks.get(taskId);
-                this.sharedWeights.forEach((weights, layer) => {
-                    for (let i = 0; i < weights.length; i++) {
-                        weights[i] -= 0.01 * task.weight * gradients[i % gradients.length];
-                    }
-                });
-            });
-        }
-        const duration = Date.now() - startTime;
-        const metrics = {
-            loss: totalLoss / Math.max(1, experiencesProcessed),
-            experiencesProcessed,
-            duration,
-            taskLosses: Object.fromEntries(taskLosses),
-        };
-        return metrics;
-    }
-    /**
-     * Get task statistics
-     */
-    getTaskStats() {
-        return Array.from(this.tasks.values()).map(task => ({
-            id: task.id,
-            name: task.name,
-            performance: task.performance,
-            weight: task.weight,
-            priority: task.priority,
-        }));
-    }
-    /**
-     * Get task relationship matrix
-     */
-    getTaskRelationships() {
-        const relationships = new Map();
-        this.tasks.forEach((_, task1Id) => {
-            const similarities = new Map();
-            this.tasks.forEach((_, task2Id) => {
-                if (task1Id !== task2Id) {
-                    similarities.set(task2Id, this.computeTaskSimilarity(task1Id, task2Id));
-                }
-            });
-            relationships.set(task1Id, similarities);
-        });
-        return relationships;
-    }
-    /**
-     * Select action for specific task
-     */
-    async selectActionForTask(state, taskId, context) {
-        const prediction = await this.forwardMultiTask(state, taskId);
-        return {
-            id: String(Math.floor(Math.abs(prediction) * 10)),
-            embedding: Array.isArray(state) ? state : [state],
-            confidence: Math.abs(prediction),
-            metadata: { taskId },
-        };
-    }
-}

package/dist/plugins/implementations/neural-architecture-search.d.ts

@@ -1,148 +0,0 @@
-/**
- * Neural Architecture Search (NAS) Plugin
- *
- * Implements automated neural architecture optimization using
- * evolutionary algorithms and reinforcement learning.
- *
- * Key features:
- * - Architecture encoding and mutation
- * - Population-based search
- * - Performance prediction
- * - Efficient architecture evaluation
- * - Transfer learning from meta-architectures
- */
-import { BasePlugin } from '../base-plugin';
-import { Action, Context, Experience, TrainOptions, TrainingMetrics } from '../learning-plugin.interface';
-/**
- * Layer specification
- */
-interface LayerSpec {
-    type: 'dense' | 'conv' | 'pool' | 'dropout' | 'batchnorm';
-    size?: number;
-    activation?: 'relu' | 'tanh' | 'sigmoid' | 'leaky_relu';
-    dropout?: number;
-    kernelSize?: number;
-}
-/**
- * Architecture specification
- */
-interface Architecture {
-    id: string;
-    layers: LayerSpec[];
-    fitness: number;
-    complexity: number;
-    generation: number;
-}
-/**
- * Neural Architecture Search Plugin Implementation
- */
-export declare class NeuralArchitectureSearchPlugin extends BasePlugin {
-    name: string;
-    version: string;
-    private experiences;
-    private strategy;
-    private population;
-    private populationSize;
-    private eliteSize;
-    private mutationRate;
-    private crossoverRate;
-    private maxLayers;
-    private maxNodesPerLayer;
-    private layerTypes;
-    private bestArchitecture;
-    private generation;
-    private evaluationHistory;
-    constructor(config?: Partial<any>);
-    /**
-     * Override to skip initialization check for in-memory operation
-     */
-    protected checkInitialized(): void;
-    /**
-     * Override to store experiences in-memory without vectorDB
-     */
-    /**
-     * Override selectAction to provide base implementation
-     */
-    selectAction(state: number[] | any, context?: Context): Promise<Action>;
-    storeExperience(experience: Experience): Promise<void>;
-    /**
-     * Override to retrieve from local experiences
-     */
-    retrieveSimilar(state: number[], k: number): Promise<import('../..').SearchResult<Experience>[]>;
-    /**
-     * Initialize random population
-     */
-    private initializePopulation;
-    /**
-     * Generate random architecture
-     */
-    private generateRandomArchitecture;
-    /**
-     * Generate random layer
-     */
-    private generateRandomLayer;
-    /**
-     * Generate unique ID for architecture
-     */
-    private generateArchitectureId;
-    /**
-     * Calculate architecture complexity
-     */
-    private calculateComplexity;
-    /**
-     * Evaluate architecture fitness
-     */
-    private evaluateArchitecture;
-    /**
-     * Estimate architecture performance (simplified)
-     */
-    private estimatePerformance;
-    /**
-     * Mutate architecture
-     */
-    private mutate;
-    /**
-     * Crossover two architectures
-     */
-    private crossover;
-    /**
-     * Select parent using tournament selection
-     */
-    private selectParent;
-    /**
-     * Evolve population for one generation
-     */
-    private evolveGeneration;
-    /**
-     * Train using NAS
-     */
-    train(options?: TrainOptions): Promise<TrainingMetrics>;
-    /**
-     * Calculate population diversity
-     */
-    private calculatePopulationDiversity;
-    /**
-     * Get best architecture
-     */
-    getBestArchitecture(): Architecture | null;
-    /**
-     * Export architecture as JSON
-     */
-    exportArchitecture(arch: Architecture): string;
-    /**
-     * Import architecture from JSON
-     */
-    importArchitecture(json: string): Architecture;
-    /**
-     * Get search statistics
-     */
-    getSearchStats(): {
-        generation: number;
-        evaluations: number;
-        bestFitness: number;
-        populationDiversity: number;
-        avgComplexity: number;
-    };
-}
-export {};
-//# sourceMappingURL=neural-architecture-search.d.ts.map

package/dist/plugins/implementations/neural-architecture-search.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"neural-architecture-search.d.ts","sourceRoot":"","sources":["../../../src/plugins/implementations/neural-architecture-search.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAC5C,OAAO,EACL,MAAM,EACN,OAAO,EACP,UAAU,EACV,YAAY,EACZ,eAAe,EAEhB,MAAM,8BAA8B,CAAC;AAEtC;;GAEG;AACH,UAAU,SAAS;IACjB,IAAI,EAAE,OAAO,GAAG,MAAM,GAAG,MAAM,GAAG,SAAS,GAAG,WAAW,CAAC;IAC1D,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,GAAG,YAAY,CAAC;IACxD,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED;;GAEG;AACH,UAAU,YAAY;IACpB,EAAE,EAAE,MAAM,CAAC;IACX,MAAM,EAAE,SAAS,EAAE,CAAC;IACpB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;CACpB;AAWD;;GAEG;AACH,qBAAa,8BAA+B,SAAQ,UAAU;IAC5D,IAAI,SAAgC;IACpC,OAAO,SAAW;IAElB,OAAO,CAAC,WAAW,CAAoB;IACvC,OAAO,CAAC,QAAQ,CAAkC;IAClD,OAAO,CAAC,UAAU,CAAsB;IACxC,OAAO,CAAC,cAAc,CAAc;IACpC,OAAO,CAAC,SAAS,CAAa;IAC9B,OAAO,CAAC,YAAY,CAAe;IACnC,OAAO,CAAC,aAAa,CAAe;IAGpC,OAAO,CAAC,SAAS,CAAc;IAC/B,OAAO,CAAC,gBAAgB,CAAe;IACvC,OAAO,CAAC,UAAU,CAAkE;IAGpF,OAAO,CAAC,gBAAgB,CAA6B;IACrD,OAAO,CAAC,UAAU,CAAa;IAC/B,OAAO,CAAC,iBAAiB,CAAkC;gBAE/C,MAAM,CAAC,EAAE,OAAO,CAAC,GAAG,CAAC;IAgBjC;;OAEG;IACH,SAAS,CAAC,gBAAgB,IAAI,IAAI;IAIlC;;OAEG;IACH;;OAEG;IACG,YAAY,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,GAAG,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC;IAavE,eAAe,CAAC,UAAU,EAAE,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC;IAI5D;;OAEG;IACG,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,OAAO,EAAE,YAAY,CAAC,UAAU,CAAC,EAAE,CAAC;IAStG;;OAEG;IACH,OAAO,CAAC,oBAAoB;IAM5B;;OAEG;IACH,OAAO,CAAC,0BAA0B;IAkBlC;;OAEG;IACH,OAAO,CAAC,mBAAmB;IA+B3B;;OAEG;IACH,OAAO,CAAC,sBAAsB;IAQ9B;;OAEG;IACH,OAAO,CAAC,mBAAmB;IAyB3B;;OAEG;YACW,oBAAoB;IAgBlC;;OAEG;YACW,mBAAmB;IA6BjC;;OAEG;IACH,OAAO,CAAC,MAAM;IA4Bd;;OAEG;IACH,OAAO,CAAC,SAAS;IAiBjB;;OAEG;IACH,OAAO,CAAC,YAAY;IAapB;;OAEG;YACW,gBAAgB;IA+C9B;;OAEG;IACG,KAAK,CAAC,OAAO,CAAC,EAAE,YAAY,GAAG,OAAO,CAAC,eAAe,CAAC;IAqB7D;;OAEG;IACH,OAAO,CAAC,4BAA4B;IAKpC;;OAEG;IACH,mBAAmB,IAAI,YAAY,GAAG,IAAI;IAI1C;;OAEG;IACH,kBAAkB,CAAC,IAAI,EAAE,YAAY,GAAG,MAAM;IAU9C;;OAEG;IACH,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,YAAY;IAI9C;;OAEG;IACH,cAAc,IAAI;QAChB,UAAU,EAAE,MAAM,CAAC;QACnB,WAAW,EAAE,MAAM,CAAC;QACpB,WAAW,EAAE,MAAM,CAAC;QACpB,mBAAmB,EAAE,MAAM,CAAC;QAC5B,aAAa,EAAE,MAAM,CAAC;KACvB;CAYF"}