@astermind/astermind-elm/dist/tasks/RefinerELM.d.ts

JavaScript Extreme Learning Machine (ELM) library for browser and Node.js.

import { Activation } from '../core/ELMConfig';
type MetricsGate = {
    rmse?: number;
    mae?: number;
    accuracy?: number;
    f1?: number;
    crossEntropy?: number;
    r2?: number;
};
export interface RefinerELMOptions {
    /** REQUIRED: input vector length for numeric mode */
    inputSize: number;
    /** REQUIRED: hidden units for the ELM */
    hiddenUnits: number;
    /** Optional activation (defaults to 'relu') */
    activation?: Activation;
    /** Optional initial categories; can be overridden on train() */
    categories?: string[];
    /** Optional logging */
    log?: {
        modelName?: string;
        verbose?: boolean;
        toFile?: boolean;
        level?: 'info' | 'debug';
    };
    /** Optional export name */
    exportFileName?: string;
    /** Optional regularization / init knobs */
    ridgeLambda?: number;
    dropout?: number;
    weightInit?: 'uniform' | 'xavier' | 'he';
    /** Optional metric thresholds (set on the ELM instance, not in config) */
    metrics?: MetricsGate;
}
export declare class RefinerELM {
    private elm;
    constructor(opts: RefinerELMOptions);
    /** Train from feature vectors + string labels. */
    train(inputs: number[][], labels: string[], opts?: {
        reuseWeights?: boolean;
        sampleWeights?: number[];
        categories?: string[];
    }): void;
    /** Full probability vector aligned to `this.elm.categories`. */
    predictProbaFromVector(vec: number[]): number[];
    /** Top-K predictions ({label, prob}) for a single vector. */
    predict(vec: number[], topK?: number): Array<{
        label: string;
        prob: number;
    }>;
    /** Batch top-K predictions for an array of vectors. */
    predictBatch(vectors: number[][], topK?: number): Array<Array<{
        label: string;
        prob: number;
    }>>;
    /** Hidden-layer embedding(s) — useful for chaining. */
    embed(vec: number[]): number[];
    embedBatch(vectors: number[][]): number[][];
    /** Persistence passthroughs */
    loadModelFromJSON(json: string): void;
    saveModelAsJSONFile(filename?: string): void;
}
export {};