add_vertex_covariates to associate mesh vertices
with covariates (hean Anderson)vertex_formula, while also allowing geometric
anisotropy.triangle_formula which
permits an offset for barrier covariatesspatial_cor to use sparse matrices to compute
the correlation between a coord and pred
coordinates, e.g. to visualize covariate-based anisotropymake_eof_ram to work as intended for multivariate
models, e.g., having a separate response map and shared indices across
variables (previously had a shared response map and shared indices,
i.e., collapsed to univariate model)predict when
spatial_varying involved a column of newdata
that was expecting a factor, but provided a
character-vector. Now coercers to factors using the levels
of the original data (same behavior as for
formula)fit$rep$negloglik_i as log-likelihood for each
datum, for use in calculating out-of-sample predictive scoremu_g in predictproject when
spatial_varying was specifiedweights argument using
family of delta_gamma or delta_lognormal,
where this bug was introduced in release 1.2.0 (h/t Peri Gerson for
identifying the issue)project functionsf::st_make_gridsf::st_make_gridweights as number of
Binomial trials Ndeviance_explained to work with non-default
weights argumentfit$internal$packageVersion to allow
predict, cAIC etc to check and throw error if
there’s a package inconsistency between object and installed
packageconditional_gmrf to do conditional simulations from
a GMRFproject to project tinyVAST forward in time... argument to tinyVAST so that
it’s obvious if an argument is mis-named.devtools::check_win_devel “array subscript ’const
__m128i[0]’ is partly outside array bounds of ‘unsigned char [12]’”term_covariance to calculate the covariance
among variables for SEM term, or covariance among variables-and-lags for
DSEM termsalphaprime_j and alphaprime2_j to
fit$rep output.cv to calculate
crossvalidation skillbias.correct option to predict (but still no flag
for SEs for anything except p_i)sample_variable from using
obj$env$MC to obj$env$spHess(random=TRUE)
which seems more stable as dependencyte and ti splines,
although they remain poorly testedsfnetwork_mesh to detect if the
stream network is not ordered as a treesimulate_sfnetwork to use thattinyVAST.cpp to use matrix notation constructor
and fix bug in previous constructor where the covariance between first
and second nodes was not righttest-sfnetworks.R integrated test to confirm
that matrix-notation precision constructor is identical to the inverse
of Ornstein-Uhlenbeck covariance as intended.simulate.tinyVAST and
sample_variable to try to avoid terminal output giving
error in valgrind checkivector_minus_one function to satisfy
clang-UBSANGMRF(Q).Quadform(x) to
x.matrix().transpose() * (Q * x.matrix()) to avoid
calculating log-determinant of Q in the smoothers penalty,
to avoid valgrind errorssem to
space_term, dsem to
spacetime_term and spatial_graph to
spatial_domain, and eliminating delta_ in the
names for arguments to delta_optionstime_term to allow time-variable interaction (e.g.,
AR1 intercepts)simulate S3 genericpredict(fit, what="mu_g") and a Poisson-linked delta
modelintegrate_outputcAIC (but disabling EDF calculation for now)data has a factor with extra
levels, which conflicted with the logic of adding all
origdata levels to newdata when calling
predict, and hence caused an uniformative error
previouslyintegrate_output interface by splitting
W_gz and V_gz into four vectors
area, type, covariate, and
weighting_index to simplify documentations and improve
namingpredict and
integrate_outputdeviance_explained and calculating this by
defaultsdmTMB as dependency, and importing family
options from it