Skip to contents

Changelog

Source: NEWS.md

rwig 0.3.0

Performance work driven by perf profiles; all solver outputs are checked against frozen outputs of version 0.2.0 (relative differences below 1e-11) and the Julia-derived reference solutions in the tests.

  • wdl()/wig() on the CPU now predict the document barycenters with the same batched Gibbs-kernel iteration the training uses, on a whole batch of documents at once. Before, inference solved one barycenter per document and, with the default method = "auto", did so in the log domain, which cost about 20 times the whole training (2000 NYT headlines, 1 epoch: 33 minutes, now 33 seconds). method, threshold and n_threads of the barycenter control are documented as unused by WDL.
  • The log-stabilized sinkhorn() and barycenter() exponentiate with a vectorised rational approximation (1-2 ulp of exp()) instead of one libm call per matrix entry, and log sinkhorn() no longer computes a column soft-min per iteration whose only use, the column term of the error, is identically zero. 1000 x 1000 log Sinkhorn, 1000 iterations: 19 s to 8 s on one thread.
  • Worker threads (n_threads) spin briefly between the short parallel sections of an iteration instead of sleeping; 12 threads now give about 5x on the forward passes where they gave 3x.
  • The backward passes of the log-stabilized sinkhorn() and barycenter() (with_grad = TRUE) no longer materialize the M x N softmax matrices: the forward pass keeps its row and column soft-mins (M + N numbers per iteration), from which each adjoint product is one fused sweep over the cost matrix, split over the threads by column blocks. 1000 x 1000 log Sinkhorn with gradient, 1000 iterations: 15.6 s to 13.2 s on one thread, 3.8 s to 2.1 s on 12; 800 x 800 log barycenter with gradients: 37.9 s to 33.9 s and 10.9 s to 5.2 s.
  • Kernel products with up to 4 right-hand columns (the “parallel” barycenter()) use one dsymv/dgemv per column rather than a level-3 call that repacks the N x N kernel every time: 800 x 800 barycenter with 4 sources, 0.79 s to 0.34 s.
  • CUDA wdl(): the per-document cublasDgemv launches of the weight adjoint are one strided-batched GEMM (2000 headlines: 5.6 s to 4.5 s).

rwig 0.2.0

CRAN release: 2026-09-20

  • Dropped the dependencies on Rcpp and RcppArmadillo. The C++ code talks to R through the native C API (.Call with registered routines) and calls the BLAS/LAPACK libraries shipped with R directly through a small internal matrix layer, which also makes the installed package much smaller. The package now has no compiled-code dependencies at all.
  • Interrupting a long computation (Ctrl-C) now releases worker threads and buffers before returning to R, and surfaces as an R error.
  • Faster log-stabilized sinkhorn() and barycenter(): the soft-min kernels no longer materialize the M x N matrix C - f 1' - 1 g' on every iteration, and worker threads (n_threads) are created once per call instead of once per iteration.
  • Faster wdl() on CPU: the softmax Jacobian is applied in O(N) instead of O(N^2) per column, an unused matrix product per backward step was removed, and the per-batch history buffers use about half the memory.
  • wdl() now draws its random initialization from R’s RNG (rnorm()), so set.seed() reproduces the documented R sequence. Fits with the same seed therefore differ from version 0.1.0, which used Armadillo’s own generator.
  • Fixed: wdl() returned weights, docs_pred and docs_dist in the shuffled training order (the default shuffle = TRUE) next to docs in the input order, so wig() summed document scores into the wrong periods. Per-document outputs are now put back into the input order.
  • Fixed: wig() failed when wig_control was given without group_unit.
  • wdl_specs()/wig_specs() now honor the values passed in barycenter_control (method, max_iter, …) instead of silently overriding them; the wdl_control seed entry is named seed (it was documented as rng_seed but read as seed). verbose defaults to 0 when omitted from a partial sinkhorn_control/barycenter_control.
  • Dropped the dependency on RhpcBLASctl: rwig no longer sets the BLAS thread count to 1 for the whole session when attached. The startup message now explains the recommendation and, if RhpcBLASctl is installed (it is now only suggested), reports the current thread count and the call to change it.
  • Dropped the dependency on lubridate: group_unit is now passed to cut() (?cut.Date), so it accepts “day”, “week”, “month”, “quarter”, “year” and multiples such as “2 months”. Weeks start on Monday.
  • Fixed: threaded (n_threads > 0) log barycenter crashed when the cost matrix had more columns than rows.
  • Fixed: the CUDA build passes an explicit GPU architecture to nvcc (recent CUDA toolkits default to sm_52, which lacks double-precision atomicAdd). Set RWIG_CUDA_ARCH to override the detected flag, or RWIG_NO_CUDA=1 to build without GPU support even when CUDA is installed.

rwig 0.1.0

CRAN release: 2026-04-21