/**
 * @license
 * Copyright 2021 Google LLC. All Rights Reserved.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 * https://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 * =============================================================================
 */
import * as tf from '@tensorflow/tfjs';
import type { CustomCallbackArgs } from '@tensorflow/tfjs';
import { CustomHandPose, type Metadata } from './custom-handpose';
export interface TrainingParameters {
    denseUnits: number;
    epochs: number;
    learningRate: number;
    batchSize: number;
}
export declare class TeachableHandPose extends CustomHandPose {
    private trainDataset;
    private validationDataset;
    private __stopTrainingResolve?;
    /** Array of training examples per class, each element is a class bucket */
    examples: Float32Array[][];
    private seed?;
    /** True once the model has been trained (more than 2 layers built) */
    get isTrained(): boolean;
    get isPrepared(): boolean;
    get numClasses(): number;
    /**
     * Add a sample of data under the provided class index.
     * @param className Integer index for the class this example belongs to
     * @param sample Feature vector from `handOutputsToArray()`
     */
    addExample(className: number, sample: Float32Array): void;
    /**
     * Classify a hand feature vector using the trained model.
     * Returns full probability distribution across all classes.
     */
    predict(handOutput: Float32Array): Promise<any[]>;
    /**
     * Classify a hand feature vector and return the top-K predictions.
     */
    predictTopK(handOutput: Float32Array, maxPredictions?: number): Promise<any[]>;
    /**
     * Pre-process collected examples into train / validation tf.data.Datasets.
     * Must be called before `train()` or will be called automatically.
     */
    prepare(): void;
    private convertToTfDataset;
    /**
     * Build and train the Dense classifier head on the collected examples.
     *
     * Architecture:
     *   Dense(denseUnits, relu) → BatchNorm → Dropout(0.3)
     *   → Dense(denseUnits/2, relu) → BatchNorm → Dropout(0.2)
     *   → Dense(numClasses, softmax)
     *
     * @param params Training hyper-parameters
     * @param callbacks Keras-compatible training callbacks
     */
    train(params: TrainingParameters, callbacks?: CustomCallbackArgs): Promise<tf.LayersModel>;
    /**
     * Save the trained model to a given IO handler or URL.
     */
    save(handlerOrURL: tf.io.IOHandler | string, config?: tf.io.SaveConfig): Promise<tf.io.SaveResult>;
    /** Initialise the examples array (one bucket per class). Must be called after setLabels(). */
    prepareDataset(): void;
    stopTraining(): Promise<void>;
    dispose(): void;
    setLabel(index: number, label: string): void;
    setLabels(labels: string[]): void;
    getLabel(index: number): string;
    getLabels(): string[];
    setName(name: string): void;
    getName(): string | undefined;
    /** Optional seed for reproducible data shuffling. */
    setSeed(seed: string): void;
    /**
     * Calculate per-class accuracy on the held-out validation set.
     * Returns tensors with predicted and ground-truth class indices.
     */
    calculateAccuracyPerClass(): Promise<{
        reference: tf.Tensor<tf.Rank>;
        predictions: tf.Tensor<tf.Rank>;
    }>;
}
/**
 * Create a new TeachableHandPose instance ready for training.
 * @param metadata Partial metadata (labels can be added later via setLabels)
 */
export declare function createTeachable(metadata: Partial<Metadata>): Promise<TeachableHandPose>;