//::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
/** @author  John Miller
 *  @version 1.6
 *  @date    Mon Sep  2 16:24:38 EDT 2013
 *  @see     LICENSE (MIT style license file).
 *
 *  @title   Data Reduction: Base Trait for Dimensionality Reducers
 */

package scalation.analytics

import scalation.linalgebra.MatriD

//::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
/** The `Reducer` trait provides a common framework for several data reduction
 *  algorithms.
 */
trait Reducer
{
    //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
    /** Reduce the original data matrix by producing a lower dimensionality matrix
     *  that maintains most of the descriptive power of the original matrix.
     */
    def reduce (): MatriD

    //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
    /** Reduce the original data matrix by factoring it into two lower dimensionality
     *  matrices that maintains most of the descriptive power of the original matrix.
     *  Override to algorithms that use factoring.
     *  @see `NMFactortorization`
     */
    def factorReduce (): (MatriD, MatriD) = (reduce (), null)

    //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
    /** Approximately recover the original matrix.  The new matrix will have
     *  the same dimensionality, but will have some lose of information.
     */
    def recover (): MatriD

} // Reducer trait