combine() now aligns named estimates and their
information matrices by parameter name. Incompatible, duplicate,
missing, or partially named parameters are rejected rather than silently
pooling different quantities. Fully unnamed estimates retain positional
matching.combine(list()) now reports that at least one estimate
is required.rmap() now forwards transformation arguments to the
transformed point estimate as well as to the variance calculation, for
both delta and Monte Carlo methods.pred() using point estimate instead of random
draw for first Monte Carlo sample, breaking the parameter-uncertainty
integrationconfint() ignoring the parm
argument. Now honors parameter subsetting by name or index, matching the
stats::confint contract.confint() producing NaN without informative error
when variance-covariance has negative diagonal (bad Hessian). Now gives
actionable error message.combine() giving cryptic “non-conformable arrays”
error when MLEs have different parameter dimensions. Now validates
upfront.sampler.mle_fit_boot() unreachable code branch
(length(x$t) == 1 is never true for boot objects; corrected
to ncol(x$t) == 1L)stud case in
confint.mle_fit_boot() switchpoint() references in roxygen and vignette
prose (now params())bias.mle_normal vignette example missing
... in method signatureS3 class renamed from "mle" to
"mle_fit" to resolve name collision with
stats4::mle (S4 class). Subclasses follow:
"mle_fit_numerical", "mle_fit_boot",
"mle_fit_rmap". Constructor function names
(mle(), mle_numerical(),
mle_boot()) are unchanged.
Removed aic(), bic(), and
loglik_val() generics. Use standard R generics
AIC(), BIC(), and logLik()
instead. logLik() returns a proper "logLik"
object with df and nobs attributes, so
AIC() and BIC() work automatically via
stats::AIC.default.
Added coef.mle_fit() method delegating to
params() for standard R compatibility.
Removed mle_weighted() constructor and
"mle_weighted" class. Use combine() instead —
same inverse-variance weighting with better error handling and a
variadic API.
1:n patterns with seq_len()
throughout to avoid edge-case bugspred()"_PACKAGE"
sentinel)fixing/ directory (removed dead experimental
code)n parameter in
rmap()joint() composes independent MLEs with disjoint
parameter sets into a joint MLE with block-diagonal covariance
structurecombine() optimally weights independent MLEs for the
same parameter via inverse-variance (Fisher information) weightingas_dist() converts MLE objects to their asymptotic
normal distributions, bridging to the algebraic.dist
distribution algebradensity(),
cdf(), inv_cdf(), sup(),
dim(), mean(), conditional()rmap() where
c() merged parameter names with transformation output
namescoef() S3 method for base R compatibility
(delegates to params())logLik() S3 method returning proper
logLik object with df and nobs
attributes, enabling automatic AIC() and BIC()
support from base Rrmap() to accept numeric n parameter
(previously required integer)mle) with methods for:
params,
nparams)vcov, se)confint)AIC, logLik)bias, mse)observed_fim)sampler)pred)expectation)marginal)mle_numerical) for
optim() resultsmle_boot) for small samplesrmap)mle_weighted)