inlabru 2.13.0

New features

• Allow the optional weights argument to bru_obs(family = "cp")/lgcp()
  to contain individual point observation weights, so that the eta contribution to
  the log-likelihood is sum(weights * eta) instead of sum(eta) (version
  2.12.0.9006)
• Automatically detect purely additive linear models (version 2.12.0.9014)
• Add experimental predictor aggregation helper feature
  bru_obs(..., aggregate = ..., aggregate_input = ...) to simplify
  specification of models with aggregation as the final step of the predictor
  evaluation (version 2.12.0.9013, bugfix in 2.12.0.9016). Includes support
  for constructing the aggregation information, via domain,samplers, or
  precomputed ips (version 2.12.0.9022)
• Add bru_set_missing() method for setting missing values in the
  bru_obs data, e.g. for use in cross-validation or prior sampling
  (version 2.12.0.9024)

Updates features

• Add automated support for INLA models with hidden states, beyond the "bym" and
  "bym2" models (version 2.12.0.9002)
• Add automatic mapper support for INLA lattice models "rw2d", "rw2diid", and
  "matern2d" (version 2.12.0.9004)
• Add inputs data to the bru_info object, so that the component inputs
  can be pre-evaluated before automated mapper construction, and avoiding
  duplicate evaluation in iinla(). This can halve the pre-processing
  time for large spatial and spatio-temporal models (version 2.12.0.9010)
• Add bru_log() data for warnings and errors reported by inlabru (version
  2.12.0.9011)
• Rename bru_like object class to bru_obs (version 2.12.0.9017)
  and temporarily re-reintroduce like_list() and bru_like_list() as aliases
  for bru_obs_list() (version 2.12.0.9020)
• Add quantile argument to spde.posterior() for controlling the credible
  interval calculations, like materncov.bands(), which is now also exported
  (version 2.12.0.9019)
• Remove dic and waic from default control.compute
bru_options, to match INLA defaults (version 2.12.0.9024)

New and updated mapper features

• The standard mapper class names have been shortened from bru_mapper_<type>
  to bm_<type>, to make code using them more readable. Objects of the old
  class names will be converted to the new classes internally, so that old
  stored objects will still work. Constructors of the form bru_mapper_<type>()
  call the corresponding new constructor bm_<type>() (version 2.12.0.9021)
• Add bm_sum() mapper, for automated adding the output of
  multiple mappers, optionally with a single common input (version 2.12.0.9001)
• Add interleaved option to bm_repeat() to allow interleaved
  states for summation of a repeated mapper (version 2.12.0.9001)
• Allow n_block in input argument to bm_aggregate and
  bm_logsumexp evaluation methods, overriding the optional mapper
  object setting (version 2.12.0.9005)
• Allow character block information in bm_aggregate and
  bm_logsumexp mappers, from fmesher version 0.2.0.9017
  (version 2.12.0.9013)
• Expanded auto-detection of component sizes by checking Cmatrix and graph
  arguments, if present and n is NULL (version 2.12.0.9012)
• Code refactor to expand bm_list mapper list handling, removing unnecessary
  method layers for component linearisation and simplification
  (version 2.12.0.9018)

Bugfixes and deprecations

• Give deprecation warnings for Spatial object inputs to bru_obs(), as
  maintaining the fallback support code is becoming increasingly time-consuming,
  and the sf package is now the recommended spatial data handling package.
  (version 2.12.0.9023)
• Bugfix for generate.bru() for evaluation of expressions in the absence of
  newdata (version 2.12.0.9007)
• Use mid locations for ibm_values() for non-indexed fm_mesh_1d mapper
  (version 2.12.0.9009)
• Deprecate the include and exclude arguments to predict() and
  generate() (version 2.12.0.9003)
• Assumed that NROW() on the main component gave the correct size for group
  and replicate in the component _eval() feature.
  Now uses ibm_n_output() instead. Fixes #271 (version 2.12.0.9015)