NEWS

posterior 1.6.0 (2024-07-03)

Add exclude option to subset_draws(), which can be used to exclude the matched selection.
Add are_log_weights option to pareto_smooth(), which is necessary for correct Pareto smoothing computation if the input vector consists of log weights.
Add pareto_smooth option to weight_draws(), to Pareto smooth weights before adding to a draws object.
Add individual Pareto diagnostic functions (pareto_khat(), pareto_khat_threshold(), pareto_min_ss(), pareto_convergence_rate())
thin_draws() now automatically thins draws based on ESS by default, and non-integer thinning is possible.
Matrix multiplication of rvars can now be done with the base matrix multiplication operator (%*%) instead of %**% in R >= 4.3.
variables(), variables<-(), set_variables(), and nvariables() now support a with_indices argument, which determines whether variable names are retrieved/set with ("x[1]", "x[2]" ...) or without ("x") indices (#208).
Add extract_variable_array() function to extract variables with indices into arrays of iterations x chains x any remaining dimensions (#340).
For types that support factor variables (draws_df, draws_list, and draws_rvars), extract_variable() and extract_variable_matrix() can now return factors.

Added support for nested-Rhat via rhat_nested (#256)
Added support for indexing draws in rvars using rvars (#282):
- x[i] or x[i] <- y where i is a scalar logical rvar slices (or updates) x by its draws. Thus, if y <- x[i], then y is the same shape as x but with sum(i) draws.
- x[[i]] or x[[i]] <- y where i is a scalar numeric rvar slices (or updates) x by selecting the ith element within each corresponding draw. Thus, if y <- x[[i]], then y is an rvar of length 1.
Added rvar_ifelse(), which is a variant of ifelse() that accepts (and returns) rvars (#282).
Array broadcasting for rvars has been made faster.

Ensure rfun() works with primitive functions (#290) and dots arguments (#291).
Provide implementations of vctrs::vec_proxy_equal(), vctrs::vec_proxy_compare(), and vctrs::vec_proxy_order().
Minor future-proofing of cbind(<rvar>), rbind(<rvar>), and chol(<rvar>) for R 4.4 (#304).
Ensure that bind_draws(<draws_rvars>) regenerates draw ids when binding along chains or draws; this also fixes a bug in split_chains(<draws_rvars>) (#300).

Delay applying tibble::num() formatting to output from summarise_draws() until print() is called so that summary output can be easily converted to a vanilla data frame (#275).

Added new rvar_factor() and rvar_ordered() subtypes of rvar() that work analogously to factor() and ordered() (#149). See the new section on rvar_factors in vignette("rvar").
The draws_df(), draws_list(), and draws_rvars() formats now support discrete variables stored as factors / ordereds (or rvar_factors / rvar_ordereds). If converted to formats that do not support discrete variables with named levels (draws_matrix() and draws_array()), factor-like variables are converted to numerics.
Made match() and %in% generic and added support for rvars to both functions.
Added modal_category(), entropy(), and dissent() functions for summarizing discrete draws.
Allow lists of draws objects to be passed as the first argument to bind_draws() (#253).
Improving formatting of summarise_draws output via tibble::num.
print.rvar() and format.rvar() now default to a smaller number of significant digits in more cases, including when printing in data frames. This is controlled by the new "posterior.digits" option (see help("posterior-package")).
Implemented faster vec_proxy.rvar() and vec_restore.rvar(), improving performance of rvars in tibbles (and elsewhere vctrs is used).

Ensure that as_draws_rvars() preserves dimensions of length-1 arrays (#265).
Fix some minor compatibility issues with rvar, vctrs, dplyr, and ggplot2 (#267, #269).

Implement for_each_draw(x, expr), which executes expr once for each draw of x, exposing variables in x as arrays of the shape implied by the indices in their names (#224).
Implement subset_draws(), thin_draws(), and resample_draws() for rvars (#225).
Allow weights to be optional in resample_draws() (#225).
Add an implementation of drop() for rvars.
Speed up subsetting of draws_list objects. (#229, #250)

Support remaining modes of diag() for rvars (#246).
Better parsing for named indices in as_draws_rvars(), including nested use of [, like x[y[1],2] (#243).
Allow 0-length rvars with ndraws() > 1 (#242).
Ensure 0-length rvars can be cast to draws formats (#242).
Don't treat length-1 rvars with more than 1 dimension as scalars when casting to other formats (#248).

fix hidden variables in bind_draws.draws_df when binding more than two objects thanks to Jouni Helske (#204)
fix output of pillar::glimpse() when used on a data frame containing rvars (#210)
drop "draws" and "draws_df" classes from draws_df objects if meta data columns are removed by a dplyr operation (#202)
fix output of print.draws_df() on objects with unrepaired draws (#217)
ensure variance() works properly with summarise_draws() (#219)

use matrixStats to speed up convergence functions (#190) and rvar summaries (#200)

ensure that as_draws_rvars() works on lists of lists (#192)
fix some vector recycling issues with rvar_rng (#195)
ensure that subset_draws() respects input variable order, thanks to Karl Dunkle Werner and Alexey Stukalov (#188)

No longer check for constant-per-chain input in effective sample size diagnostics as this is overly conservative especially for ess_tail. (#198)