federer/dist/src/common/dataset.d.ts

Experiments in asynchronous federated learning and decentralized learning

import * as tf from "@tensorflow/tfjs-node";
/**
 * Interface of an object that holds metadata about a {@link Dataset}.
 *
 * @typeParam T Type of the metadata
 */
export interface DatasetInfo<T> {
    train: DataSubsetInfo<T>;
    test: DataSubsetInfo<T>;
}
/**
 * Interface of an object that holds metadata about a {@link DataSubset}.
 *
 * @typeParam T Type of the metadata
 */
export interface DataSubsetInfo<T> {
    items: T;
    labels: T;
}
/** Type of an object holding file paths for a serialized {@link Dataset} */
export declare type DatasetFilepaths = DatasetInfo<string>;
/** Type of an object holding file paths for a serialized {@link DataSubset} */
export declare type DataSubsetFilepaths = DataSubsetInfo<string>;
/**
 * A dataset is a collection of data. This data is split into multiple
 * {@link DataSubset}s. For now, only one split is possible: splitting into a
 * training subset and a test subset.
 *
 * @typeParam IR The rank of the tensors holding items (IR means "Item Rank")
 * @typeParam LR The rank of the tensors holding labels (LR means "Label Rank")
 */
export declare class Dataset<IR extends tf.Rank = tf.Rank, LR extends tf.Rank = tf.Rank> {
    readonly data: {
        readonly train: DataSubset<IR, LR>;
        readonly test: DataSubset<IR, LR>;
    };
    constructor(data: {
        readonly train: DataSubset<IR, LR>;
        readonly test: DataSubset<IR, LR>;
    });
    /** Get the training subset */
    get train(): DataSubset<IR, LR>;
    /** Get the test subset */
    get test(): DataSubset<IR, LR>;
    /** Equivalent to `tf.dispose`. */
    dispose(): void;
    /**
     * Counts the number of datapoints present in the {@link Dataset}. Each
     * item and its associated label constitutes a single datapoint.
     *
     * @returns The number of datapoints in the {@link Dataset}
     */
    countNumberDatapoints(): number;
    concat(that: Dataset<IR, LR>): Dataset<IR, LR>;
    /**
     * Save the dataset to files.
     *
     * @param directory Directory to save to
     * @param filenamePrefix Prefix of the dataset filenames
     * @return A promise of the filepaths that the dataset has been saved to
     */
    save(directory: string, filenamePrefix: string): Promise<DatasetFilepaths>;
    /**
     * Load a dataset from files.
     *
     * @param filePathsOrUrls Filepaths or URLs to load the dataset from
     * @param expectedRanks Expected ranks of the items and labels in the dataset.
     * If not provided, the ranks of the loaded data will not be checked.
     * @return A promise of the loaded dataset, with the expected ranks.
     */
    static load<IR extends tf.Rank = tf.Rank, LR extends tf.Rank = tf.Rank>(filePathsOrUrls: DatasetFilepaths, expectedRanks?: {
        items?: IR;
        labels?: LR;
    }): Promise<Dataset<IR, LR>>;
    /** Cast to a {@link Dataset} with given item and label ranks. */
    withRanks<OIR extends IR, OLR extends LR>(ranks: {
        items?: OIR;
        labels?: OLR;
    }): Dataset<OIR, OLR>;
}
/**
 * A data subset is a subset of the datapoints of a {@link Dataset}.
 *
 * The {@link https://en.wikipedia.org/wiki/Training,_validation,_and_test_sets |
 * standard terminology} for this is "dataset" or "set", but this is confusing
 * because the full dataset is also called a "dataset". We therefore call this
 * a "data subset", as it is a subset of the full dataset.
 *
 * @typeParam IR The rank of the tensors holding items (IR means "Item Rank")
 * @typeParam LR The rank of the tensors holding labels (LR means "Label Rank")
 */
export declare class DataSubset<IR extends tf.Rank = tf.Rank, LR extends tf.Rank = tf.Rank> {
    readonly data: {
        readonly items: tf.Tensor<IR>;
        readonly labels: tf.Tensor<LR>;
    };
    constructor(data: {
        readonly items: tf.Tensor<IR>;
        readonly labels: tf.Tensor<LR>;
    });
    get items(): tf.Tensor<IR>;
    get labels(): tf.Tensor<LR>;
    /** Equivalent to `tf.dispose`. */
    dispose(): void;
    concat(that: DataSubset<IR, LR>): DataSubset<IR, LR>;
    /**Splits into two with split ratio  */
    split(split: number): DataSubset[];
    /**
     * Counts the number of datapoints present in the {@link DataSubset}. Each
     * item and its associated label constitutes a single datapoint.
     *
     * @returns The number of datapoints in the {@link DataSubset}
     */
    countNumberDatapoints(): number;
    /**
     * Save the data subset to files.
     *
     * @param directory Directory to save to
     * @param filenamePrefix Prefix of the data subset filenames
     * @return A promise of the filepaths that the data subset has been saved to
     */
    save(directory: string, filenamePrefix: string): Promise<DataSubsetInfo<string>>;
    /**
     * Load a data subset from files.
     *
     * @param filepathOrURL Filepaths or URLs to load the data subset from
     * @param expectedRanks Expected ranks of the items and labels in the dataset.
     * If not provided, the ranks of the loaded data will not be checked.
     * @return A promise of the loaded data subset, with the expected ranks.
     */
    static load<IR extends tf.Rank, LR extends tf.Rank>(filepathOrURL: DataSubsetInfo<string>, expectedRanks?: {
        items?: IR;
        labels?: LR;
    }): Promise<DataSubset<IR, LR>>;
    /** Cast to a {@link DataSubset} with given item and label ranks. */
    withRanks<OIR extends IR, OLR extends LR>(ranks: {
        items?: OIR;
        labels?: OLR;
    }): DataSubset<OIR, OLR>;
}
//# sourceMappingURL=dataset.d.ts.map