list
of values for the
initial
, partition
, and/or
base_optimizer
parameters, where each combination creates a
separate alternating optimization thread.Removed update_code
from output. Updates are now
accepted regardless whether they improve the function value. In the end,
the best solution is returned.
Added tolerance_history
argument to compare function
value and parameters between more than one iteration.
Added hide_warnings
argument to hide warnings during
the AO procedure.
Implemented random partitions, parameter bounds, new stopping criteria (iteration limit, time limit, change in function value, change in parameters).
The ao()
API did change a lot, please check the
documentation.
{cli}
to improve error messages.Added package start up-message.
Added more details to README file.
Increased required version of {optimizeR}
to version
1.0.3.
{optimizeR}
version 1.0.1.Renamed input print.level
->
verbose
.
In the return value, renamed element "time"
->
"seconds"
(time is now always in seconds); the same in the
"sequence"
element. Additionally, added
"value"
column to the "sequence"
element.
Added input f_partition
to optionally specify
functions for the partitions.
Added input joint_end
to optionally optimize jointly
in the end of the alternating optimization procedure.
Input iterlim
can be Inf
now.
Having empty elements in partition
is no longer
allowed.
Removed plotting feature for parameter updates during estimation.
New dependency on {checkmate}
for input
checks.
{pkgdown}
package. No
code changes.Updated code to the new {optimizeR}
version 0.3.0,
which renamed changed some input and output names.
In particular, in ao()
, renamed input
optimizer
-> base_optimizer
.
Fixed bug when using numerical gradient and/or Hessian in target function.
{optimizeR}
version 0.2.0,
which renamed some functions....
arguments for
ao()
.Removed the set_f()
interface. The optimization
problem is now specified directly in ao()
.
The output format of ao()
has been changed, see the
documentation of ao()
.
Arbitrary optimizer can now be specified in ao()
as
an optimizer
object via the framework from the {optimizeR}
package.
Removed utils functions from {ao}
. They are imported
from the {optimizeR}
package.
ao()
now returns the sequence of estimates in the
different iterations as a data frame.
ao()
now is able to plot the sequence of estimates
during estimation.
Argument groups
in ao()
now is called
partition
and sequence
is replaced by
iterations
.
New function set_f()
for specification of the
optimization problem. Parameter constraints can now be imposed on the
target function.
New helper functions is_number()
,
try_silent()
, timed()
and
euclidean()
.
Included a vignette with details on alternating optimization.
New package sticker.
Updated package metadata.
Updated package sticker.
print.ao()
.Skip alternation optimization step if a group is empty.
Estimation times now is returned always in seconds.
Return stats::nlm()
outputs.
Implemented method print.ao()
.