Releases: DylanB95/statespacer
Releases · DylanB95/statespacer
statespacer 0.5.0
statespacer 0.4.1
statespacer 0.4.1
Removed dependency
- Removes the dependency on the YieldCurve package by including an updated FedYieldCurve dataset in the statespacer package. See
?FedYieldCurve
for details.
statespacer 0.4.0
statespacer 0.4.0
Extra functionality
- Introduces the simulation smoother for drawing random samples conditional on the observed data. See
?SimSmoother
for details.
Performance improvements
- Extraction of components now done a bit more efficiently.
Bug fixes
- Fixed incrementation of initialisation_steps during initialisation when the dependent variable was missing.
statespacer 0.3.0
statespacer 0.3.0
Performance improvements
- Kalman Filter and Smoother now fully written in c++.
Extra functionality
- Computation of diagnostics now optional using
diagnostics = TRUE
. - Added full system matrices to the statespacer object.
- Added predicted, filtered, and forecasted
P_star
andP_inf
.
Bug fixes
- Fixed bug in predicting using explanatory variables.
- Fixed edge case in adding explanatory variables. The bug occurred when there was only 1 explanatory variable and 1 dependent variable supplied, causing the Z system matrix to be a vector instead of an array.
statespacer 0.2.1
statespacer 0.2.1
Bug fixes
- Patch for macOS. On macOS, assignment of nested lists is handled a bit differently than on other platforms. For instance, if
first <- list()
, then assigningfirst$second[[["third"]] <- 1
returns on Windows (and other platforms) a list namedfirst
, that contains another list namedsecond
, that contains a named elementthird
equal to an unnamed length 1 numerical vector. On macOS though, it returns a list namedfirst
, that contains a named elementsecond
equal to a named (third
) length 1 numerical vector. Sosecond
is a list on the other platforms, while being a named numerical vector on macOS. This caused a bug on macOS while computing standard errors.
statespacer 0.2.0
statespacer 0.2.0
Breaking changes
-
API change by making use of S3 classes and methods:
-
Replaced
StateSpaceFit()
forstatespacer()
. -
Replaced
StateSpaceEval(...)
forstatespacer(..., fit = FALSE)
. -
statespacer()
returns a list object of classstatespacer
. -
Replaced
StateSpaceForecast()
for the S3 methodpredict.statespacer()
.
-
Performance improvements
-
Major:
- Calculation of loglikelihood now done using c++. Major performance improvement as the loglikelihood potentially gets calculated a lot of times by the optimisation procedure.
-
Minor:
-
Making use of crossprod and tcrossprod.
-
Improved efficiency of the computation of standard_errors.
-
y_temp outside of LogLikelihood function.
-
Easier to find proper initial values, reducing time spent on trial and error by the user.
-
Extra functionality
-
Printing progress now optional using
verbose = TRUE
. -
Computation of standard_errors now optional using
standard_errors = TRUE
.
statespacer 0.1.0
statespacer 0.1.0
- Initial release to CRAN!