version 1.0.0

Changelog

@@ -0,0 +1,27 @@
+# 2023-03-23 v0.3.2 [AG]
+- Added additional methods of halflife, unconditional and pit for use with
+tsgarch and tsarma.
+
+# 2022-06-09 v0.3.1 [AG]
+- Converted documentation to Roxygen and performed some clean up.
+
+# 2022-04-15 v0.3.0 [AG]
+- Added a tsmoments method
+- Bumped the version to 0.3.0 to be consistent with major changes in other 
+packages with the same version.
+
+# 2022-03-16 v0.1.8 [AG]
+- Added a tscalibrate method
+
+# 2021-05-20 v0.1.7 [AG]
+- Added additional estimate_ad method for new plugin packages which use
+automatic differentiation.
+- Fix to tsgrowth documentation
+
+# 2021-04-20 v0.1.6 [AG]
+- Fix to prediction plots to allow missing values in observed series (ylim 
+fixed for na.rm)
+
+# 2020-11-16 v0.1.5 [AG]
+- Fix to tsreport method (previously dispatched to report rather than tsreport).
+- Added Changelog to package

DESCRIPTION

@@ -0,0 +1,21 @@
+Package: tsmethods
+Type: Package
+Title: Time Series Methods
+Version: 1.0.0
+Date: 2024-03-03
+Authors@R: c(person("Alexios", "Galanos", role = c("aut", "cre","cph"), email = "alexios@4dscape.com"))
+Maintainer: Alexios Galanos <alexios@4dscape.com>
+Description: Generic methods for use in a time series probabilistic framework, allowing for a common calling convention across packages. Additional methods for time series prediction ensembles and probabilistic plotting of predictions is included. A more detailed description is available at <https://www.nopredict.com/packages/tsmethods> which shows the currently implemented methods in the 'tsmodels' framework. 
+License: GPL-2
+Imports: methods, zoo, xts
+Encoding: UTF-8
+URL: https://www.nopredict.com/packages/tsmethods,
+        https://github.com/tsmodels/tsmethods
+RoxygenNote: 7.3.1
+Suggests: testthat (>= 3.0.0)
+Config/testthat/edition: 3
+NeedsCompilation: no
+Packaged: 2024-03-08 18:48:29 UTC; alexios
+Author: Alexios Galanos [aut, cre, cph]
+Repository: CRAN
+Date/Publication: 2024-03-09 11:00:02 UTC

MD5

@@ -0,0 +1,35 @@
+00ec9a14ec1dafa6e88fac97384182c3 *Changelog
+30e634f599938e04f6e9f00abae93b05 *DESCRIPTION
+04e96b1340c8072787791b5d31a674b5 *NAMESPACE
+e9b3821d37fd289fc1663ec88c84bd0f *R/distributions.R
+f3956cf38c5f778e1b1fcdf9410adbbb *R/methods.R
+244a7fa38f5dd4c272fe496d49f66baa *R/plots.R
+1765c53b3d5321b919bd0d5ae4679ce9 *R/tsmethods-package.R
+20f191df7cd9aa4293579c7278628b8f *man/ensemble_modelspec.Rd
+aaadf33c852b427964f50c53df01826b *man/estimate.Rd
+6deb816a4a119e64428ec9e8880d2003 *man/estimate_ad.Rd
+af26c5981e7f6d40a8b628f1170a5e8c *man/halflife.Rd
+65527ea0376d9bc83d7bd67a978c8f92 *man/pit.Rd
+0277d9954f9cc77e0454569bcbe07b10 *man/plot.Rd
+fe70876e6f9689ad736b725c372537a3 *man/tsaggregate.Rd
+266de3fede27c18807cbec8452770aa6 *man/tsbacktest.Rd
+0ed2d6bdd79a32369326e53c19e13f4b *man/tsbenchmark.Rd
+af0567566067067cad4b93afcc319c9e *man/tscalibrate.Rd
+05cc68b4b5e64a5193182ff508c221da *man/tsconvert.Rd
+48fe9175d5ed620be700bcdbd7e46716 *man/tscor.Rd
+08a3f90bc36c2246eb14efd24e55faef *man/tscov.Rd
+618b4e2c730b7eec4fcfb6bcb842854a *man/tsdecompose.Rd
+4df879080d6887229943888bbf631af3 *man/tsdiagnose.Rd
+3516f9ab61b054cf18d824543fb50da7 *man/tsensemble.Rd
+7bb176a909277b3dd1ee6a2760eb02c2 *man/tsequation.Rd
+ddf6eced4b48e41f15668c5c1fbb4252 *man/tsfilter.Rd
+a62beef78e3b6bddccfa013c9569ff14 *man/tsgrowth.Rd
+695c59bdd220d730f6a4bcd65aba57f9 *man/tsmethods-package.Rd
+80df777325c5017c2047451610ad6c59 *man/tsmetrics.Rd
+9fc89b408cd79069404c98b80cbb5211 *man/tsmoments.Rd
+e46dd55ad0503a3ec6e2d5b285cb98f5 *man/tsprofile.Rd
+f39a2ce39b84c45dae3795211c1eac15 *man/tsreport.Rd
+a76a3379713f56250eb7d479d049030c *man/tsspec.Rd
+22d150757d09360f83fa2264b341aa8e *man/unconditional.Rd
+f92b06bcdc0fd11b80ca1de9ac0a4825 *tests/testthat.R
+31cab4aa8039503411190ac41835379f *tests/testthat/test-ensemble.R

NAMESPACE

@@ -0,0 +1,63 @@
+# Generated by roxygen2: do not edit by hand
+
+S3method(plot,tsmodel.distribution)
+S3method(plot,tsmodel.predict)
+S3method(tsensemble,ensemble.spec)
+S3method(tsgrowth,tsmodel.predict)
+export(ensemble_modelspec)
+export(estimate)
+export(estimate_ad)
+export(halflife)
+export(pit)
+export(tsaggregate)
+export(tsbacktest)
+export(tsbenchmark)
+export(tscalibrate)
+export(tsconvert)
+export(tscor)
+export(tscov)
+export(tsdecompose)
+export(tsdiagnose)
+export(tsensemble)
+export(tsequation)
+export(tsfilter)
+export(tsgrowth)
+export(tsmetrics)
+export(tsmoments)
+export(tsprofile)
+export(tsreport)
+export(tsspec)
+export(unconditional)
+import(methods)
+importFrom(grDevices,col2rgb)
+importFrom(grDevices,colorRampPalette)
+importFrom(grDevices,gray)
+importFrom(grDevices,n2mfrow)
+importFrom(grDevices,rgb)
+importFrom(graphics,abline)
+importFrom(graphics,axis)
+importFrom(graphics,axis.Date)
+importFrom(graphics,axis.POSIXct)
+importFrom(graphics,box)
+importFrom(graphics,grid)
+importFrom(graphics,hist)
+importFrom(graphics,layout)
+importFrom(graphics,legend)
+importFrom(graphics,lines)
+importFrom(graphics,par)
+importFrom(graphics,plot)
+importFrom(graphics,points)
+importFrom(graphics,polygon)
+importFrom(stats,fitted)
+importFrom(stats,median)
+importFrom(stats,na.omit)
+importFrom(stats,quantile)
+importFrom(stats,ts)
+importFrom(utils,data)
+importFrom(utils,head)
+importFrom(utils,tail)
+importFrom(xts,endpoints)
+importFrom(xts,xts)
+importFrom(zoo,coredata)
+importFrom(zoo,index)
+importFrom(zoo,zoo)

R/distributions.R

@@ -0,0 +1,136 @@
+#' Ensemble specification
+#'
+#' @description Creates and validates an ensemble specification.
+#' @param ... objects of either all of class \dQuote{tsmodel.predict} or 
+#' \dQuote{tsmodel.distribution} representing the probabilistic forecasts for 
+#' the same horizon of optionally different models on the same series and with 
+#' same number of draws. It is expected that the predictive distributions are 
+#' based on joint simulated draws passed to the \code{innov} argument in 
+#' the \code{predict} function of the supporting models. Instead of \ldots 
+#' it is also possible to pass a list of the objects.
+#' @returns An object of class \dQuote{ensemble.spec}.
+#' @aliases ensemble_modelspec
+#' @rdname ensemble_modelspec
+#' @export
+#'
+#'
+ensemble_modelspec <- function(...)
+{
+    m <- list(...)
+    # if its already a list
+    if (length(m) == 1) {
+        if (is.list(m[[1]])) {
+            m <- m[[1]]
+        }
+    }
+    object_class <- sapply(1:length(m), function(i) tail(class(m[[i]]),1))
+    if (length(unique(object_class)) != 1) stop("\nThe object must have the same class (tsmodel.predict or tsmodel.distribution)")
+    if (!object_class[1] %in% c("tsmodel.predict","tsmodel.distribution")) stop("\nThe object must be of class tsmodel.predict or tsmodel.distribution")
+    object_class <- object_class[1]
+    if (object_class == "tsmodel.predict") {
+        original_series <- do.call(cbind, lapply(1:length(m), function(i) m[[i]]$original_series))
+    } else {
+        original_series <- NULL
+    }
+    n <- length(m)
+    if (object_class == "tsmodel.predict") {
+        horizon <- NCOL(m[[1]]$distribution)
+        forc_dates <- colnames(m[[1]]$distribution)
+        sims <- sapply(1:length(m), function(i) NROW(m[[i]]$distribution))
+        if (length(unique(sims)) != 1) stop("\nThe simulated distributions must have the same number of draws (rows)")
+        # check horizon match
+        check <- sapply(2:length(m), function(i){
+            all.equal(forc_dates, colnames(m[[i]]$distribution))
+        })
+        if (!all(check)) stop("\nhorizon and forecast dates of each distribution must match exactly.")
+        M <- array(0, dim = c(dim(m[[1]]$distribution), length(m)))
+        for (i in 1:length(m)) M[,,i] <- as.matrix(m[[i]]$distribution)
+    } else {
+        horizon <- NCOL(m[[1]])
+        forc_dates <- colnames(m[[1]])
+        sims <- sapply(1:length(m), function(i) NROW(m[[i]]))
+        if (length(unique(sims)) != 1) stop("\nThe simulated distributions must have the same number of draws (rows)")
+        # check horizon match
+        check <- sapply(2:length(m), function(i){
+            all.equal(forc_dates, colnames(m[[i]]))
+        })
+        if (!all(check)) stop("\nhorizon and forecast dates of each distribution must match exactly.")
+        M <- array(0, dim = c(dim(m[[1]]), length(m)))
+        for (i in 1:length(m)) M[,,i] <- as.matrix(m[[i]])
+    }
+    out <- list(distribution = M, original_series = original_series, object_class = object_class, forc_dates = forc_dates)
+    class(out) <- "ensemble.spec"
+    return(out)
+}
+
+#' @method tsensemble ensemble.spec
+#' @details Returns the weighted distribution, under the assumption that the 
+#' predictions were generated using a joint error distribution whose values were 
+#' passed to the \code{innov} argument of the predict function used for each model.
+#' @returns An object of class \dQuote{tsmodel.predict} or \dQuote{tsmodel.distribution} 
+#' depending on the input class in \code{\link{ensemble_modelspec}}.
+#' @rdname tsensemble
+#' @export
+#'
+#
+tsensemble.ensemble.spec <- function(object, weights = NULL, ...)
+{
+    n <- dim(object$distribution)[3]
+    if (is.null(weights)) {
+        weights <- matrix(rep(1/n, n), ncol = 1)
+        warning("\nweights are NULL. Setting to equal weight ensemble")
+    } else {
+        if (length(as.numeric(weights)) != n) stop("\nlength of weights must be equal to number of distributions passed.")
+        weights <- matrix(weights, ncol = 1)
+    }
+    d <- t(apply(object$distribution, 1, function(x) x %*% weights))
+    colnames(d) <- object$forc_dates
+    class(d) <- "tsmodel.distribution"
+    # ensemble the original series if present
+    if (object$object_class == "tsmodel.predict") {
+        ind1 <- index(object$original_series)
+        original_series <- xts(rowSums(coredata(object$original_series) * matrix(weights, ncol = n, nrow = nrow(object$original_series), byrow = TRUE), na.rm = TRUE), ind1)
+        L <- list(distribution = d, original_series = original_series, mean = zoo(colMeans(d), as.Date(object$forc_dates)))
+        class(L) <- "tsmodel.predict"
+    } else {
+        L <- d
+        class(L) <- "tsmodel.distribution"
+    }
+    return(L)
+}
+
+
+#' @param object an object.
+#' @param d the period back to look at for growth calculations.
+#' @param type the type of growth calculation. \dQuote{diff} is simply the 
+#' difference in values over n periods, \dQuote{simple} if the 
+#' rate of change and \dQuote{log} the difference in logs.
+#' @param ... additional parameters passed to the method.
+#' @returns an object of class \dQuote{tsmodel.predict} transformed to a growth distribution.
+#' @method tsgrowth tsmodel.predict
+#' @rdname tsgrowth
+#' @export
+#'
+#
+tsgrowth.tsmodel.predict <- function(object, d = 1, type = c("diff","simple","log"), ...)
+{
+    Z <- matrix(coredata(object$original_series), ncol = length(object$original_series), nrow = nrow(object$distribution), byrow = TRUE)
+    colnames(Z) <- as.character(index(object$original_series))
+    G <- cbind(Z, as.matrix(object$distribution))
+    if (type == "diff") {
+        P <- G[,(d + 1):ncol(G)] - G[,1:(ncol(G) - d)]
+    } else if (type == "simple") {
+        P <- G[,(d + 1):ncol(G)]/G[,1:(ncol(G) - d)] - 1
+    } else if (type == "log") {
+        P <- log(G)
+        P <- P[,(d + 1):ncol(P)] - P[,1:(ncol(P) - d)]
+    }
+    N <- NCOL(object$distribution)
+    PN <- NCOL(P)
+    D <- P[,tail(1:PN,N)]
+    class(D) <- "tsmodel.distribution"
+    original_series <- zoo(P[1,1:(PN - N)], index(object$original_series)[-(1:d)])
+    L <- list(original_series = original_series, distribution = D)
+    class(L) <- "tsmodel.predict"
+    return(L)
+}