fix problem with multicore support, now correct functions are used
from pbmcapply
add info in documentation and warning that on Windows only 1 core is
used
add toc to readme
fastpos 0.5.0
improve multicore support, pbapply is now used to split up the
calculation of points of stability; this is much faster than with
futures but does not work on Windows
give the option to set the corridor manually via lower_limit and
upper_limit, which will override precision parameters
change parameter precision to precision_absolute and precision_rel
to precision_relative, if you used these parameters before, you need to
change your code
parameter rhos is no rho
precision_relative now takes a numeric value, instead of being a
logical
change how corridor limits are set up, now closures are used to
handle all cases
several improvements in documentation, adding more details and
hints
several improvements in the readme, especially for the
benchmarks
linting the package with lintr
deprecate old precision parameters (preparing for version 0.6.0) and
rhos
improve parameter order
use scientific notation more consistently
use roxygen md
fastpos 0.4.1
tests now show more information if they fail, also the tolerance for
an individual deviation is higher to avoid false alarms
progress bars are only shown if R is run interactively (this is
required as per the manual for writing R extensions)
fastpos 0.4.0
multiple cores are now supported via the future package, use
parameter n_cores, gains are only visible if
n_studies is high (overhead)
tests for multiple cores were included
replace parameter was added to the externally visible
functions, and tests for replace = false
minor improvements in layout/documentation
test added for old pop_create function
test coverage is now somewhat higher since the untestable proportion
of code reduced
fastpos 0.3.0
Parameter for number of studies is now always n_studies. Some lower
level functions used number_of_studies.
create_pop now creates a bivariate normal distribution with an exact
rho, this is slower than previously, but it is probably worth it.
Problems with C++ progressbar (RcppProgress) were fixed. The code
should also be cleaner now.
Make C++ and Rstudio interruption similar: depending on the timing,
either Rstudio interrupts by itself and stops everything or C++ returns
a value. In the latter case, R now also stops quietly.
Since the interrupts cannot be properly tested, code coverage is
only 90%.
fastpos 0.2.0
If the corridor of stability is not reached, NA is returned by an
internal function that simulate_pos build upon.
n_not_breached is simply the number of NA values. The maximum
sample size is still used for the calculation of the quantiles if the
corridor was not reached, which should be better than ignoring the
specific study altogether (i.e. treating it as an NA value).
simulate_pos now returns an IntegerVector (instead of
NumericVector) to better handle NA values.
A simple test for relative precision was added. Another test was
added that unloads the package. Code coverage is now 100%.
The vignette and readme were improved substantially. Word choice
and grammar were checked by an editor. A section on the speed of
fastpos in comparison to corEvol was added including a
theoretical argument and an empirical test.