legion v0.1.2 (Release data: 2023-01-30)
=======
Changes:
* The order in which in plot.legion() now matters.
* Allow estimating seasonal vets() on the short data (one season of data).
* auto.vets() now does not check for common initial level and common level component. This is not practical.
Bugfixes:
* Fixed a bug in dimnames for some vets() models.
* ves() and vets() would produce an error in case of h=1 due to drop of one of the dimensions by R, when extracting the holdout.
* Fix in auto.vets() for level models with no checks for initials and components.
* measures() was not imported from greybox for some reason. Now it is.
* Different initialisation of persistence matrix in ves(), so that it does not get stuck in the violation of stability condition.
legion v0.1.1 (Release data: 2022-02-14)
=======
Changes:
* Usual bounds are now available in vets. This will make all parameters lie between 0 and 1.
* Custom loss function for both ves() and vets().
* We now import forecast from generics package, not from greybox.
* Correct names of states that include original names of time series in vets().
Bugfixes:
* forecast.legion would not work correctly if variables did not have names.
legion v0.1.0 (Release data: 2021-05-15)
=======
Changes:
* Imported ves() and vets() functions from smooth package.
* We now have explicit lags parameter in ves() and vets().
* First working version of vets().
* architectorVETS in vets() - internal function needed for selector.
* vets() and ves() now support model selection.
* vets() now has a proper likelihood estimation and implementation of multiplicative error model.
* ves() now also has a proper MLE.
* Parameter "silent" can now only be TRUE or FALSE.
* auto.vets() function, selecting PIC elements via I->P->C.
* plot() method for ves() and vets(). Currently only produces plots 1 and 4-7.
* Fixed BICc and AICc for multivariate models.
* loss="diagonal" now refers to likelihood maximisation, assuming that the series are independent.
* auto.vets() now starts with the model with loss="diagonal", implying that covariances are all zero.
* Substituted the parameters "y" with "data".
* A brief vignette for vets().
* plot.legion() method now supports all plots.
* forecast.legion() and related methods is now available.
* ves() and vets() now work with zoo classes of data.
* on.exit() restore par for plot functions.