now the plot methods for time series objects are exported, so they work again (they had stopped working due to changes in R 4.0).
corrections of typo’s and other minor tweaks in the documentation.
vastly improved support for dates/times.
new generic pcIntercept.
updated methods for pcMean.
pc_sdfactor now returns a matrix also in the case maxlag = 0 (as documented).
pc.sdfactor was renamed to pc_sdfactor. pc.sdfactoris available but deprecated.
new function pc_mean.
new dataset Fraser2017.
pc_sum gets argument na.rm and a more efficient implementation.
new methods for zoo::na.trim.
new subset PAR models with trigonometric parameterisation.
now using some datetime functions from lubridate. Also reexportinglubridate::dateandlubridate::date<-`.
first draft of a data vignette.
added missing predict, residuals, and fitted methods.
improved pcTest().
included “nsadata.csv” in the package, currently as internal data for testing.
Need description and maybe a better name before exporting. Object datansa is the whole dataset, object nsaauto is column “AUTOMOTIVEPRODNSA”.
removed deprecated function ptildeorders(), use pdSafeParOrder() instead.
removed class union “AnyTimeSeries”, it had not been in use for a long time.
fixed use of suggested package to comply with CRAN policies
in test_piar(), now p-values are set to NA if package fUnitRoots is not available and a message is issued. Previously an error was thrown.
removed package pear from dependencies.
consolidated the dependencies.
replaced wrong use of is.na() with gbutils::isNA() or other suitable code.
some old code contained instances of class(x) == something, now fixed.
changed slightly some Rd files not rendered well by pkgdown, e.g. move commented items out of ‘describe’ in methods’ descriptions.
extensive testing and bug fixing.
fixed a bug revealed by changes in R-devel circa start of February 2020. It was related to a change in variable names for some interactions produced by lm(), see this R-devel question: https://r.789695.n4.nabble.com/changed-names-from-lm-in-R-devel-td4761450.html).
edited README.md to reflect the CRAN release of the package, since it contained some text that was meant for the pre-CRAN Github version only.
exportClass to export classes rather than exportClassPattern with a lazy regexp.added an example for pclspiar
edited DESCRIPTION to comply with CRAN policies.
removed deprecated functions mCpar and sim_arAcf.
prepared for release.
replaced setIs, introduced in v0.12-0 for "PeriodicAutocovariances" and related classes, with direct inheritance.
numerous other changes, consolidations and bug-fixes. In particlar, updated periodic integrated and seasonally integrated models, including fitPM().
now require ‘lagged (>= 2.2)’ (for the new "slMatrix" method for [[).
now require ‘R (>= 3.5.0)’ (for isFALSE). Before release require ‘R (>= 3.6.0)’ - there is no point to cater for old versions of R, given that this is a complete rework.
partialAutocorrelations() for the new class ‘PartialPeriodicAutocorrelations’ class return by default the variances in lag 0. This was initially the case years ago but then I changed that in some cases. The convention with the variances in lag 0 keeps the full information, so are equivalent parameterisation. TODO: Methods for printing and plotting may present the them differently.
now the first argument of fitPM() is model (swapped the places of the first two arguments to achieve that).
consolidated PAR methods for fitPM() to use the periodic time series classes.
for fitPM(), before the periodic time series classes it was convenient to consider a matrix, m as representing a periodic time series with nrow(m) seasons (see pcMatrix()). Now dropping this convention in fitPM() - for now throw an error if x is a matrix with more than one column (in future could consider fitting multivariate PAR).
wrapping up v0.11-0 before starting changes to the model classes and related functionality. The reference manual and the rendered org files printed on 13 May 2019 are after the final changes to v0.11-0 (except this note and the date in DESCRIPTION).
now the periodic time series classes PeriodicTS and PeriodicMTS have more or less complete functionality. Methods are defined for a number of generics from R base related to time series.
added slot pcstart to class Cyclic and did a number of adjustements to the periodic time series classes to use this.
redesigned pcCycle(), its API was too convoluted.
new function BultinCycle() creates instances from the builtin classes.
now class "BultinCycle" is virtual.
consolidated time series anc cycle classes.
registered periodic time series methods (S3) for stats::monthplot() and stats::boxplot().
first version developed under git. I created the git repo starting with the earliest available version and adding all available versions one by one (one commit for each version). Finally I added and committed the curent development directory and did some clean-up. Before starting any work under the repo, I changed the version number to 0.9-4 and made a commit. The master branch was the only branch at that time.
fixed a bug in initialize() for class "SimpleCycle" - the logical in an if could be of length larger than one for multi-season nSeasons(). This is flagged as an error in R-deve for R 2.7-0 (April 2019).
small edits of the documentation.
R CMD check --as-cran passes completely.
consolidated the autocovariance classes and acf computations.
some bug fixes and other improvements.
moved ‘slMatrix’ class and function to package ‘lagged’.
moved also sl2acfbase(), acfbase2sl() and sl2vecacf()to package ‘lagged’ (file acfutils.R).
renamed sim_arAcf() to sim_parAcf().
fixed a bug in intercept2permean().
consolidated the names of some simulation functions (for now the old names work but are deprecated.
moving packages ‘methods’, ‘Matrix’, ‘mcompanion’ from Depends to Imports (‘sarima’ stays in Depends, at least for now; maybe this is its natural place).
turned NEWS into NEWS.md.
removed unnecessary (and unused) argument lag_0 from filterPoly() methods for classes “PeriodicBJFilter” and “PeriodicSPFilter”.
Package working again; R CMD check completes with no WARNINGs and NOTEs. (With --as-cran gives two NOTEs: - that mcompanion is not on CRAN, - non-standard files in the root directory).
Revamped the model classes.
The time series classes are as in Version 0.7-0.
A number of R files are now generated from ’*.org’ sources (this was started in 0.7-0).
pc.xxx functionsnSeasons since it doesn’t make sense.R CMD check without any warnings/notes.pc.maxlag() to maxLag().pc.nseasons() is now obsolete - it is no longer imported from pcData and is defined to call nSeasons().partial consolidation of “slMatrix” - the code was from 2006-2007 with occasional patches.
mass renaming with more consistent names (most of the code is from 2007 or earlier when underscore was not admissible in names),
new pc time series classes
new argument ‘nseasons’ for pc_test; renamed argument ‘model’ to ‘nullmodel’.
renamed pc_test to pcTest.
renamed pcAcf to pcAcvf - acf is universally used for autocorrelations
now import “zoo”.
now import “ltsa” (McLeod et al).
many other changes - new classes, consolidations.
changing to new version before making changes to eliminate package pcData.
pclsdf was giving warning about length of residuals non-multiple of number of seasons, when computing innovation variances.
formula construction in pclsdf somewhat cleaned up.
there are now classes for fitted models.
num2pcpar now returns a list if ‘result’ is NULL and PAR coefficients otherwise (used to look at argument ‘mean’ to decide on this, now gives error if ‘mean’ is invalid).
matcovlist now has argument “result” and can return a matrix.
Create a new package pctsData and move some stuff there to reduce the clutter in “pcts”. For now make “pcts” depend on “pctsData” but the idea is eventually to drop this dependance, since much of the stuff that will be put there is redundant. (In fact, eventually the dependance may be the other way round (“pctsData” depending on “pcts” for those needing compatibility).
modified xx.ss and similar to use subspace parameterisation for core vectors. Partially implemented. Works with version 0.2-6 of “mcompanion”.
xx.ss; corresponding changes in mC.ss.xx.ss - more meaningful processing of initial values.pcls.R.pc.armafilter and pc.filter (long overdue). Existing calls to pc.filter.arma can simply be replaced by calls to pc.filter.xarma. Calls to pc.armafilter can be replaced by calls to pc.filter but in addition argument ‘model’ should be named since in pc.filter it is the first argument, while in pc.armafilter it was third.started updating the time series classes.
bug in pc.filter: failed to change sign of intercept and nintercept when whiten = TRUE, which was giving wrong results in the case of non-zero intercepts.
removed other bugs