random_sevt and random_parentslist.depsubtree which build the dependency subtree.sevt and ceg objects to graphs (edge lists and igraph).plot.sevt and on graph conversion methods when ignoring stage “1” which is assigned to rootsevt_fit earns the scope argument, it is now possible to do partial re-fit of staged tree models. Moreover, with the compute_logLik argument, the user can decide if the log-likelihood should be computed when fitting the model.stages_hclust can now search the best cut of the hierarchical cluster with respect to some score function to be maximized.stages_simplebhc implements a new stage structure search by iterative joining of positions, thus searching in the space of simple staged trees.sevt_simplify return the simplified staged tree, that is a stage tree where positions and stages are equivalent.ceg when model has less than two variables.cli package for errors and warnings.sevt objects.stages function and new indexing of stages via a new dedicated class sevt.stgs.stages_csbhc, a new stages learning algorithm which iteratively add context specific independencies. Additionally functions ci_matrices and join_all have been added.inst/CITATION file added as requested by JSS.plot.ceg new plotting functions using igraph plotting.make_stages_col which help computing stages colors for sevt and ceg plotting.sample_from now returns a data.frame of factors.prob earns a new argument conditional_on, that makes easier to compute conditional probabilities.confint.sevt, implement a method for confidence intervals for the parameters of a model of class sevt.lr_test new function, implementing likelihood-ratio test.search_best and search_greedy.cid function that implements context intervention discrepancy.sevt and DAG representations.as_sevt.bn.fit.find_stage.summary.sevt for objects without ctables.path_probability and prob for sevt objects with not ordered $prob.fix color assignment consistency between plot.sevt and barplot.sevt
This version introduces major changes, in functions capabilities and in functions naming. These changes are almost surely breaking any previous code using older versions of the package. In particular, all functions named *.sevt but class methods are now called differently. Moreover, various improvements and functionalities are added to better deal with unobserved situations and to improve computations. Additional model selection methods based on clustering are now available.
COMPLETE CHANGELOG:
fit parameter from full, indep. Now full and indep always fit the model while sevt is just the basic constructor of the sevt class.full and indep by default unobserved situations are joined using join_unobserved, and probabilities are fitted only after the unobserved situations are joined, improving speed. Moreover, the name of the unobserved stages are stored as name_unobserved in the staged tree object.new_label to improve speed.plot.sevt allows now to set edges color with col_edges.plot.sevt and barplot.sevt it is possible to specify stages that should be ignored and not plotted via the ignore argument, by default the name_unobserved stages are ignored.plot.sevt adds variables names by default (var_names argument).compare_stages: because of changes in plot.sevt we need to specify that the root is always considered identical.stndnaming accepts now uniq, prefix and ignore arguments, which control how stage names are generated and if some stage names should be left untouched (default: the name_unobserved stages).stages_bj (previously bj.sevt) distance is now passed with a character and no longer as a function.stages_hclust and stages_kmeans, to learn stage structure using hierarchical or k-means clustering.scope and ignore parameters that allow to specify among which variables run the algorithm and which stages should be left untouched (default: the name_unobserved stages).1:length(x) with the suggested seq_along in all code.stages_bj and stages_hclust are compatibles.as.sevt, only implemented one method for bn.fit class from bnlearn packageinclusions_stages and provide better output.summary, stages were wrongly matched to probabilitiesbarplot_stages to draw barplots of the floret probabilities. Implemented relative tests.plot.sevt.order can be passed to staged_ev_tree.bn.fitjoin_zero alias for join_zero_countsindep, probabilities should not be of class table (it was triggering a bug in summary)subtree, now removing unused probabilities (it was triggering a bug in summary)get_stage and get_path and relative testsxor is renamed as noisy_xornoisy_xorNaN log-probabilities in predict.sevtsummary.sevt function addedsubtree.sevt works now for fitted objectsinclusion.stages functionpredict.sevt now returns conditional probabilities P(C|X)predict.sevt works also if newdata does not include the classcompare.sevt works with three different methods.join_stages throws error with bad stages.lambda = 0 does not induce errors in logLik, join_stages and thus in the model selection algorithms.nvar.sevt bugceg.sevtvarnames.sevt and nvar.sevtfit.sevt is replaced by sevt.fit (following the lm.fit naming). fit.sevt is still available but marked as (DEPRECATED) in doc.compare.sevt is updated with info on the capabilities.First released version of the package