- Bug: if x0 had a column name (which it would if the data had column names), then the line to get the parvec would not have
x0.x0
as the name. It would havex0
and this would cause a fail of the check that the parvec names match whatMARSSvectorizeparams()
expects.
- Add check for data class in dfaTMB.
- Removed the kronecker products from the for loops to speed things up
- Don't create A x TT matrices when only need A x 1 because parameter is not time-varying
- Bug in
MARSS_TMB.R
that would not add one the obj when control$trace = 2.
- Big clean. Move out uni.hpp (and R), estimate_marxss.hpp (and R), MARSS_tmb.R
- Keep dfaTMB for now while testing.
- various bug fixes. Major one involving Z. Had (t(X) %x% I) instead of (I %x% t(X))
- bug with covariates since did not pass in form="marss" to coef()
- added
marss2.hpp
andestimate_marss2.R
. These use the chol() of the var-cov matrices and don't split into diag and corrmat. Seems more stable.
Created a development version of {MARSS} with method TMB. marssTMB()
will recognize if it was called with MARSS()
. MARSS_tmb()
also still works.
- Added
marss.hpp
, a major update for MARSS models in vectorized "marss" form. This will allow time-varying parameters more easily and models with linear constraints. It's companion function isestimate_marss.R
. - Changed
MARSStmb.R
toestimate_marxss.R
. It will become an internal function later. - Added various helper functions in
inst/include/LOM.hpp
- Added a Quick Start vignette.
- Added check that no zeros on diagonal of Q or R (temporary)
- Added marssTMBCheckPackageVersions() to
zzz.R
and.onLoad()
to deal with users getting warnings about TMB/Matrix version mismatch and not knowing what to do. https://glmmtmb.github.io/glmmTMB/#glmmtmbtmbmatrix-mismatches
- updated
marxss.hpp
to be in MARSS format with X and Y as mxT and nxT. - added Q, C, U, x0 and A estimation to
marxss.hpp
. Minimal testing so far. - V0 = 0 is allowed.
- tinitx=1 or tinitx=0 allowed.
- removed
to_marssTMB()
(not needed) - completed
MARSStmb()
so that the marssMLE object is in proper form. All the {MARSS} helper functions should work. - add more time comparisons to MARSS_tmb vignette and upped the maxit for MARSS()
- Got
MARSStmb()
mostly working with marssMLE structure for output - made default optimizer nlminb (faster)
- Eric added tests and covariates to DFA vignette.
- moved MARSS into Depends and added to imports in
marssTMB-package.R
- added some more info to
MARSStmb()
description - started draft of vignette for
MARSS_tmb()
- Working on
to_marssMLE()
which will convert the output to MARSS form.
To do
- Fixing the parameter output from TMB to have the parameter names. Done 0.0.6
- Might also can add a
is.diag
flag to data if I need to id if R (and later Q) is diagonal to use faster code (diagonal matrices). probably not needed?
- Added
MARSS_tmb()
. This sets up the model forMARSStmb()
and for now ensures that the model will work withmarxss.hpp
which only allows DFA at the moment. - Added
MARSStmb()
. This is symmetric toMARSS::MARSSoptim()
. Working to match the output toMARSS::MARSS()
so that all the MARSS functions work. - Added
src/TMB/marxss.hpp
which is the MARSS model version with MARSS parameter names.
To do
- Currently
MARSStmb()
just fits a model. Next up is to convert this to MARSS output format. Done 0.0.5
The first draft with the TMB package structure in place. Used the {TMBtools} package to set-up the R package to work with TMB.
- Two functions
dfaTMB()
anduniTMB()
which I will likely combine later.