//::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
/** @author  John Miller, Hao Peng
 *  @version 1.5
 *  @date    Sat Dec  8 14:32:12 EST 2018
 *  @see     LICENSE (MIT style license file).
 */

package scalation.analytics.forecaster

import scala.collection.mutable.{LinkedHashMap, Map}
import scala.math.{min, sqrt}

import scalation.linalgebra.{VectoD, VectorD}
import scalation.math.double_exp
import scalation.plot.Plot

//::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
/** The `Forecaster` trait provides a common framework for several forecasters.
 *  Note, the 'train' method must be called first followed by 'eval'.
 *  @param t       the time vector
 *  @param y       the input vector (time series data)
 *  @param orders  the orders for the model (e.g., ARIMA (,d,)) 
 */
abstract class ForecasterVec (t: VectoD, y: VectoD, orders: Int*)
         extends Forecaster
{
    protected val n  = y.dim                            // size of the input vector
    protected val ml = min (n, 20)                      // maximum lag to consider
    protected val mu = y.mean                           // the sample mean
    protected var e  = new VectorD (n)                  // residual/error vector [e_0, e_1, ... e_m-1]

    //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
    /** Given a time series 'y', train the prediction function 'y = f(y_)', where
     *  'f(y_)' is a function of the lagged values of 'y', by fitting its parameters.
     */
    def train (): ForecasterVec

    //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
    /** Return the parameter vector.
     */
    def parameters: VectoD

    //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
    /** Compute the error and useful diagnostics for the entire dataset.
     */
    def eval () = { diagnose (y, e) }

    //::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
    /** Plot a function, e.g.,  Auto-Correlation Function 'ACF', Partial Auto-Correlation
     *  Function 'PACF'.
     *  @param fVec  the vector given function values
     *  @param name  the name of the function
     */
    def plotFunc (fVec: VectoD, name: String)
    {
        val lag_axis = new VectorD (ml+1)
        for (i <- 0 until fVec.dim) lag_axis(i) = i
        val zero = new VectorD (ml+1)
        new Plot (lag_axis, fVec, zero, "Plot of " + name)
    } // plotFunc

} // ForecasterVec abstract class