v2.0.3

v2.0.3

Features

• Release to CRAN.

• tMIIC version for temporal causal discovery on stationary time series:
  new mode of miic() to reconstruct networks from temporal stationary
  datasets (Simon et al., eLife 2024).
  The temporal mode of miic() is not activated by default and can be enabled by
  setting the newly added parameter mode to "TS"(Temporal Stationary).
  A tuning of the temporal mode is possible through a set of new parameters:
  max_nodes, n_layers, delta_t, mov_avg and keep_max_data.

• Addition of the 'is consequence' prior knowledge. Consequence variables are
  excluded from the possible contributors, edges between consequences are
  ignored and edges between a non consequence and a consequence are pre-oriented
  toward the consequence.
  Information about consequence variables can be provided to miic()
  in the state_order, by supplying an is_consequence column.

• iMIIC version introducing contextual variables, genuine vs putative causes
  and multiple enhancements to deal with very large datasets (Ribeiro-Dantas et al., iScience 2024).
  Information on contextual variables can be provided to miic()
  in the state_order, by supplying an is_contextualcolumn and
  genuine vs putative causes can be tuned by the newly added parameter
  ort_consensus_ratio.

• Enhancement of orientations using mutual information supremum principle for
  finite datasets (Cabeli et al., Why21 at NeurIPS 2021).
  The use of enhanced orientations is controlled by the newly added parameter
  negative_info of miic() and is activated by default.

Fixes and improvements

• Faster post-processing in R for datasets with large number of variables.

• Fix for memory overflow on shared memory space.

• The discretization of continuous variables has been improved when dealing
  with variables having a large number of identical values.

Breaking changes

As part of this major release, consolidation of long-pending breaking changes:

• Harmonization of exported function names using camel case.

• Harmonization of parameters and return values using snake case.

• Harmonization of abbreviations.

All the documentation has been updated accordingly, in case of issue when
upgrading to this version, please consult the help of the relevant function
for more information about its interface.

For the core miic() function, the main breaking changes in the interface
when upgrading from the 1.5.3 release are:

in the parameters:

• cplx: renaming of the complexity term "mdl" → "bic"

• ori_proba_ratio → ort_proba_ratio

in the miic object returned:

• all.edges.summary → summary

  • Nxy_ai → n_xy_ai
  • log_confidence → info_shifted
  • infOrt → ort_inferred
  • trueOrt → ort_ground_truth
  • isOrtOk → is_inference_correct
  • isCausal → is_causal
  • proba → p_y2x, p_x2y
  • consensus → ort_consensus

• orientations.prob → triples

  • NI3 → ni3
  • Error → conflict

Still compared to 1.5.3, another important change in the behavior of miic()
is that, by default, miic() no longer propagates orientations
and allows latent variables discovery during orientation step.

Known issues

• Conditioning on a (very) large number of contributors can lead to a memory
  fault.