| NEWS | R Documentation |
Fixed incorrect use of data_file() functions in the documentation, pointed out by Michael Lai.
Added function data_matrix() to support data from a sparse
matrix object in R, contributed by David Cortes
(@david-cortes,
#17).
Fixed the different result on macOS.
Fixed the dependence on rmarkdown, thanks to Kurt Hornik.
Fixed the problem that recosystem gives different results on Windows compared with other OS, reported by Prof. Michael Trusov.
Updated LIBMF to the newest version (403153c) in the upstream repository.
Fixed a logical operation bug and a virtual destructor bug detected by CRAN.
Added the missing "nbin" parameter in $train() and
$tune(). (#6,
reported by @Tutuchan).
Added a progress bar for model tuning, controlled by the "progress"
parameter in $tune().
(#5,
suggested by @talegari).
Fixed the misleading column name "rmse" in the return value of
$tune().
Also updated the documentation accordingly.
Fixed the wrong parameter name in .Call(), thanks to Kurt Hornik.
Now using the prettydoc package to format vignettes.
Updated LIBMF to version 2.01.
API change from LIBMF 2.01:
The cost option in $train() and $tune()
has been expanded to and replaced by costp_l1,
costp_l2, costq_l1, and costq_l2, to
allow for more flexibility of the model.
A new loss parameter in $train() and
$tune() to specify loss function.
Other API change:
Data input and output are now managed in a unified way
via functions data_file(), data_memory(),
out_file(), out_memory(),
and out_nothing(). See ?data_source and
?output_format for more information.
As a result, a number of arguments in functions
$tune(), $train(), $output(),
and $predict() now should be objects returned by these
input/output functions.
Updated LIBMF to version 1.2.
Due to API change in LIBMF 1.2, $convert_train() and
$convert_test() have been removed, and $train()
and $predict() have changed API accordingly.
New member function $tune() for tuning model parameters.
New member function $output() to output factorization matrices.
(Internal) Avoided using C RNG.
Fixed UBSAN runtime errors, thanks to Prof. Brian Ripley.
(Internal) Using pthreads on Windows to support multithreading.
(Internal) Using portable code to allocate aligned memory.
Rolled back to Rcpp due to a NOTE in R CMD check.
Changed package name from Recosystem to recosystem.
Now using R to generate random numbers in the algorithm, so the result is reproducible by setting seeds inside R.
Now linking to Rcpp11.
A new parameter verbose to control output information.
Initial version.