Handling of dependent parameter domains: These should be specified in the
parameter domain definition and, for now, only numerical parameter can
define dependent domains. A numerical domain can be dependent on one bound,
e.g. (1, "param1*2")
, where the dependent bound can include basic
arithmetic operators. (Leslie Pérez Cáceres, Manuel López-Ibáñez)
The package now provides an ablation
executable (ablation.exe
in
Windows) that makes easier to perform ablation analysis without having any R
knowledge.
The interface to functions ablation()
and plotAblation()
has been
simplified. The ablation()
function now allows overriding scenario
settings. The plotAblation()
function will not create the plot if the
ablation log does not contain a complete ablation.
(Manuel López-Ibáñez)
The argument n.instances
of ablation()
has been renamed to n_instances
and it is now a factor that multiplies scenario$firstTest
.
(Manuel López-Ibáñez)
New command-line option --quiet
to run without producing any output
except errors (also available as a scenario option).
(Manuel López-Ibáñez)
New command-line option --init
to initialize a scenario. (Deyao Chen)
Added support for HTCondor cluster framework to --batchmode
.
(Filippo Bistaffa)
--check
now also check the contents of configurationsFile
and runs
configurations provided via initConfigurations
.
(Manuel López-Ibáñez, reported by Andreea Avramescu)
New scenario options targetRunnerLauncher
and targetRunnerLauncherArgs
to help in cases where the target-runner must be invoked via another
software with particular options (such as python.exe
in Windows).
(Manuel López-Ibáñez)
New scenario option minMeasurableTime
.
(Manuel López-Ibáñez)
An error is produced if a variable set in the scenario file is not known to
irace. If your scenario file contains R code, then use variable names
beginning with a dot '.'
, which will be ignored by irace.
(Manuel López-Ibáñez)
Plotting functions have been moved to the new package
iraceplot. In particular,
configurationsBoxplot()
is replaced by iraceplot::boxplot_training()
and
iraceplot::boxplot_test()
; parallelCoordinatesPlot()
is replaced by
iraceplot::parallel_cat()
and iraceplot::parallel_coord()
; and
parameterFrequency()
is replaced by iraceplot::sampling_frequency()
.
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
The user-guide now contains a detailed section on "Hyper-parameter optimization of machine learning methods". (Manuel López-Ibáñez)
When testType="F-test"
and only two configurations remain, the elimination
test now uses the pseudo-median estimated by the Wilcoxon signed-rank test
to decide which configuration is the best one instead of comparing the
median difference.
(Manuel López-Ibáñez)
New functions testing_fromlog()
and testing_fromfile()
for independently
executing the testing phase. The function testing.main()
was removed as it
is superseded by the new ones.
(Manuel López-Ibáñez)
New function read_logfile()
to easily read the log file produced by irace.
(Manuel López-Ibáñez)
New function printParameters()
that prints a parameters R object as a valid input text.
(Manuel López-Ibáñez)
irace2pyimp
moved to its own R package.
(Manuel López-Ibáñez)
Generating the file irace.Rdata
may be disabled by setting logFile=""
.
(Manuel López-Ibáñez, reported by Johann Dreo)
path_rel2abs()
and checkParameters()
are now exported so that other
packages may use them.
(Manuel López-Ibáñez)
path_rel2abs()
also searches in system paths. (Manuel López-Ibáñez)
readConfigurationsFile()
will now detect duplicated configurations and
error. (Manuel López-Ibáñez)
The interface to functions getFinalElites()
, getConfigurationById()
and
getConfigurationByIteration()
has been simplified.
The package provides a irace.sindef
file that may be used for building a
standalone container of irace using Singularity. See the README.md
file
for instructions. (Contributed by Johann Dreo)
New example examples/target-runner-python/target-runner-python-win.bat
contributed by Levi Ribeiro.
New helper script in bin/parallel-irace-slurm
to launch irace
in SLURM computer clusters.
(Manuel López-Ibáñez)
Rename scenario.update.paths()
to scenario_update_paths()
. The old name is deprecated. (Manuel López-Ibáñez)
Correctly handle clear out-performance cases despite strong bi-modality. (Reported by Nguyen Dang, fixed by Manuel López-Ibáñez)
Fix error when recovering from a parallel run on Windows. (Manuel López-Ibáñez, reported by Tarek Gamal)
testNbElites
now controls how many iteration elites are tested when
testIterationElites=1
. This is the documented behavior in the user guide.
(Manuel López-Ibáñez, reported by Marcelo de Souza)
Fixes to the Matlab example. (Manuel López-Ibáñez)
The default of testType
is now set to t-test
when capping is enabled.
(Manuel López-Ibáñez, reported by Jovana Radjenovic)
Fix various issues in the user guide. (Manuel López-Ibáñez, reported by Jovana Radjenovic)
Remove duplicated elites. (Manuel López-Ibáñez, reported by Federico Naldini)
Fix (#7): warnings with partial matched parameters. (Manuel López-Ibáñez, reported by Marc Becker)
Fix (#10): wrong assert with elitist=0
. (Manuel López-Ibáñez)
Fix (#12): irace can be run with FastR.
Fix (#13): Maximum number configurations immediately rejected reached. (Manuel López-Ibáñez)
Fix: when setting the scenario file in the command-line, scenarioFile
was
not set correctly. The correct scenario was used, however, the debug output
and the value stored in the log / recovery file was wrong.
(Manuel López-Ibáñez, reported by Richard Schoonhoven)
With sampleInstances = FALSE
, elitist irace does not change the order of
instances already seen. However, if you want to make sure that the order of
the instances is enforced, you also need to set elitistNewInstances=0
.
The function irace.usage()
was removed. It was not really useful for R
users as the same result can be obtained by calling
irace.cmdline("--help")
.
(Manuel López-Ibáñez)
NEWS
converted to markdown.
Fix CRAN error on Solaris.
irace2pyimp
function and executable (irace2pyimp.exe
on Windows) to
convert .Rdata files generated by irace to the input files required by the
parameter importance analysis tool PyImp
(https://github.com/automl/ParameterImportance).
(Nguyen Dang, Manuel López-Ibáñez)
Initial configurations may also be provided directly in R using
scenario$initConfigurations
(Manuel López-Ibáñez)
Rdata files are saved in version 2 to keep compatibility with older R versions. (Manuel López-Ibáñez)
Fix invalid assert with ordered parameters: (Leslie Pérez Cáceres)
value >= 1L && value <= length(possibleValues) is not TRUE
The irace
executable (irace.exe
on Windows) is a compiled binary instead
of a script. On Windows, irace.exe
replaces irace.bat
(Manuel López-Ibáñez)
inst/examples/Spear
contains the Spear (SAT solver) configuration scenario.
(Manuel López-Ibáñez)
Fixed bug when reporting minimum maxTime
required.
(Reported by Luciana Salete Buriol,
fixed by Manuel López-Ibáñez)
Fixed bug detected by assert:
R
all(apply(!is.na(elite.data$experiments), 1, any)) is not TRUE
(Reported by Maxim Buzdalov, fixed by Manuel López-Ibáñez)
Fix buggy test that breaks CRAN. (Manuel López-Ibáñez)
Do not print "23:59:59" when wall-clock time is actually close to zero. (Manuel López-Ibáñez)
Fix irace --check --parallel 2
on Windows. (Manuel López-Ibáñez)
Values of real-valued parameter are now printed with sufficient precision to
satisfy digits
(up to digits=15
).
(Manuel López-Ibáñez)
It is possible to specify boundMax
without capping.
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
irace --check
will exit with code 1 if the check is unsuccessful
(Manuel López-Ibáñez)
Print where irace is installed with --help
. (Manuel López-Ibáñez)
irace will now complain if the output of target-runner
or target-evaluator
contains extra lines even if the first line of output is correct. This is to
avoid parsing the wrong output. Unfortunately, this may break setups that
relied on this behavior. The solution is to only print the output that irace
expects.
(Manuel López-Ibáñez)
Completely re-implement log
parameters to fix several bugs. Domains that
contain zero or negative values are now rejected.
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
New option aclib=
(--aclib 1
) enables compatibility with the
GenericWrapper4AC (https://github.com/automl/GenericWrapper4AC/) used by
AClib (http://aclib.net/). This is EXPERIMENTAL. --aclib 1
also sets
digits to 15 for compatibility with AClib defaults.
(Manuel López-Ibáñez)
Fix printing of output when capping is enabled. (Manuel López-Ibáñez)
checkTargetFiles()
(--check
) samples an instance unless
sampleInstances
is FALSE. (Manuel López-Ibáñez)
Fix symbol printed in elimination test. (Manuel López-Ibáñez)
Use dynGet()
to find targetRunner
and targetEvaluator
.
As a result, we now require R >= 3.2.
(Manuel López-Ibáñez)
All tests now use testthat
. (Manuel López-Ibáñez)
New function scenario.update.paths()
(Manuel López-Ibáñez)
Fix assert failure that may happen when elitistNewInstances
is larger than
firstTest
. Reported by Jose Riveaux. (Manuel López-Ibáñez)
Fix bug in checkTargetFiles()
(--check
) with capping.
(Leslie Pérez Cáceres)
Clarify a few errors/warnings when maxTime > 0
.
(Manuel López-Ibáñez, suggested by Haroldo Gambini Santos)
Use testthat for unit testing. (Manuel López-Ibáñez)
Allow instances to be a list of arbitrary R objects (mlr
bugfix).
(Manuel López-Ibáñez)
irace now supports adaptive capping for computation time minimization.
The default value of the testType
option is t-test when adaptive capping
is enabled. Please see the user-guide for details.
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
The package contains an ablation()
function implementing the ablation
method for parameter importance analysis by Fawcett and Hoos (2016).
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
New option postselection
executes a post-selection race.
(Leslie Pérez Cáceres)
At the end of each race, if the race stops before evaluating all instances seen in previous races, then the best overall may be different than the best of the race. We now print the best overall (best-so-far). Elites evaluated on more instances are considered better than those evaluated on fewer. (Manuel López-Ibáñez, Leslie Pérez Cáceres)
Last active parameter values of numerical parameters (i
and r
) are carried
by the sampling model. When a value must be assigned and the parameter was
previously not active, the sampling is performed around the last value.
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
R help pages are now generated with Roxygen2. (Leslie Pérez Cáceres, Manuel López-Ibáñez)
The user guide documents --version
, --help
, and --check
.
(Manuel López-Ibáñez)
A return value of Inf
from targetRunner
/targetEvaluation
results in
the immediate rejection of the configuration without any further evaluation.
This is useful for handling unreliable or broken configurations that should
not stop irace. (Manuel López-Ibáñez)
Numerical parameters may be sampled on a logarithmic scale using i,log
or r,log
. (Alberto Franzin)
New target-runner.bat
for Windows contributed by André de Souza Andrade.
Fixed all shell scripts calling functions before defining them, which is not portable. (Manuel López-Ibáñez)
Fixed --parallel
bug in Windows that resulted in
Error in checkForRemoteErrors(val)
.
(Manuel López-Ibáñez)
Improve error message when no training instances are given. (Manuel López-Ibáñez)
The output of irace now specifies in which order, if any, configurations are printed. (Manuel López-Ibáñez, suggested by Markus Wagner)
Several fixes for handling paths in Windows. (Manuel López-Ibáñez)
readConfigurationsFile()
now has a text=
argument, which allows reading
configurations from a string.
(Manuel López-Ibáñez)
User-provided functions (targetRunner, targetEvaluator and repairConfiguration) and user-provided conditions for forbidden configurations are now byte-compiled when read, which should make their evaluation noticeably faster. (Manuel López-Ibáñez)
The argument 'experiment'
passed to the R function targetRunner
does not
contain anymore an element 'extra.params'
. Similarly, the 'scenario'
structure does not contain anymore the elements 'instances.extra.params'
and
'testInstances.extra.params'
. Any instance-specific parameters values now
form part of the character string that defines an instance and it is up to
the user-defined targetRunner
to parse them appropriately. These changes
make no difference when targetRunner is an external script, or when
instances and instance-specific parameter values are read from a file.
(Manuel López-Ibáñez)
iraceResults$experimentLog
to count calls to
targetEvaluator
as experiments, even if no call to targetRunner
was
performed. This does not affect the computation of the budget consumed and,
thus, it does not affect the termination criteria of irace. The bug triggers
an assertion that terminates irace, thus no run that was successful with
version 2.2 is affected.
(Manuel López-Ibáñez)Command-line parameters are printed to stdout (useful for future replications). (Manuel López-Ibáñez, suggested by Markus Wagner)
Users may provide a function to repair configurations before being evaluated. See the scenario variable repairConfiguration. (Manuel López-Ibáñez)
The option --sge-cluster
(sgeCluster
) was removed and replaced by
--batchmode
(batchmode
). It is now the responsibility of the target-runner
to parse the output of the batch job submission command (e.g., qsub
or
squeue
), and return just the job ID. Values supported are: "sge", "torque",
"pbs" and "slurm". (Manuel López-Ibáñez)
The option --parallel
can now be combined with --batchmode
to limit the
number of jobs submitted by irace at once. This may be useful in batch
clusters that have a small queue of jobs.
(Manuel López-Ibáñez)
New examples under inst/examples/batchmode-cluster/
.
(Manuel López-Ibáñez)
It is now possible to include scenario definition files from other scenario files by using:
R
eval.parent(source("scenario-common.txt", chdir = TRUE, local = TRUE))
This feature is VERY experimental and the syntax is likely to change in the future. (Manuel López-Ibáñez)
Fix a bug that re-executed elite results under some circumstances. (Leslie Pérez Cáceres)
Restrict the number of maximum configurations per race to 1024. (Leslie Pérez Cáceres)
Do not warn if the last line in the instance file does not terminate with a newline. (Manuel López-Ibáñez)
Fix bug when deterministic == 1
.
(Manuel López-Ibáñez, Leslie Pérez Cáceres)
Update manual and vignette with details about the expected arguments and
return value of targetRunner
and targetEvaluator
. (Manuel López-Ibáñez)
Many updates to the User Guide vignette. (Manuel López-Ibáñez)
Fix \dontrun
example in irace-package.Rd
(Manuel López-Ibáñez)
Fix bug: If testInstances contains duplicates, results of testing are not
correctly saved in iraceResults$testing$experiments
nor reported correctly
at the end of a run. Now unique IDs of the form 1t, 2t, ...
are used for
each testing instance. These IDs are used for the rownames of
iraceResults$testing$experiments
and the names of the
scenario$testInstances
and iraceResults$testing$seeds
vectors. (Manuel López-Ibáñez)
Fix bug where irace keeps retrying the target-runner
call even if it
succeeds. (Manuel López-Ibáñez)
New command-line parameter
--only-test FILE
which just evaluates the configurations given in FILE on the testing instances defined by the scenario. Useful if you decide on the testing instances only after running irace. (Manuel López-Ibáñez)
maxTime != 0
, the number of experiments performed may be
miscounted in some cases. (Manuel López-Ibáñez)Fix CRAN errors in tests. (Manuel López-Ibáñez)
Avoid generating too many configurations at once if the initial time estimation is too small. (Manuel López-Ibáñez)
Minimum R version is 2.15.
Elitist irace by default, it can be disabled with parameter --elitist 0
.
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
The parameter --test-type
gains two additional values: (Manuel López-Ibáñez)
t-test-bonferroni
(t-test with Bonferroni's correction for multiple
comparisons),
t-test-holm
(t-test with Holm's correction for multiple comparisons)
MPI does not create log files with --debug-level 0
.
(Manuel López-Ibáñez)
For simplicity, the parallel-irace-*
scripts do not use an auxiliary
tune-main
script. For customizing them, make a copy and edit them
directly.
(Manuel López-Ibáñez)
New parameters: (Manuel López-Ibáñez)
--target-runner-retries : Retry target-runner this many times in case of error.
We print diversity measures after evaluating on each instance: (Leslie Pérez Cáceres)
Kendall's W (also known as Kendall's coefficient of concordance) If 1, all candidates have ranked in the same order in all instances. If 0, the ranking of each candidate on each instance is essentially random.
W = Friedman / (m * (k-1))
Spearman's rho: average (Spearman) correlation coefficient computed on the ranks of all pairs of raters. If there are no repeated data values, a perfect Spearman correlation of +1 or -1 occurs when each of the variables is a perfect monotone function of the other.
Many internal and external interfaces have changed. For example, now we consistently use 'scenario' to denote the settings passed to irace and 'configuration' instead of 'candidate' to denote the parameter settings passed to the target algorithm. Other changes are:
parameters$boundary -> parameters$domain
hookRun -> targetRunner
hookEvaluate -> targetEvaluator
tune-conf -> scenario.txt
instanceDir -> trainInstancesDir
instanceFile -> trainInstancesFile
testInstanceDir -> testInstancesDir
testInstanceFile -> testInstancesFile
Minimal example of configuring a MATLAB program (thanks to Esteban Diaz Leiva)
Paths to files or directories given in the scenario file are relative to the
scenario file (except for --log-file
, which is an output file and it is
relative to --exec-dir
). Paths given in the command-line are relative to the
current working directory. Given
$ cat scenario/scenario.txt
targetRunner <- "./target-runner"
$ irace -s scenario/scenario.txt
irace will search for "./scenario/target-runner"
, but given
$ irace -s scenario/scenario.txt --target-runner ./target-runner
irace will search for "./target-runner"
. (Manuel López-Ibáñez)
New command-line wrapper for Windows installed at
system.file("bin/irace.bat", package="irace")
(thanks to Anthony Antoun)
Budget can be specified as maximum time (maxTime
, --max-time
) consumed by
the target algorithm. See the documentation for the details about how this
is handled.
(Leslie Pérez Cáceres, Manuel López-Ibáñez)
The best configurations found, either at the end or at each iteration of an
irace run, can now be applied to a set of test instances different from the
training instances. See options testInstanceDir
, testInstanceFile
,
testNbElites
, and testIterationElites
. (Leslie Pérez Cáceres, Manuel López-Ibáñez)
The R interfaces of hookRun
, hookEvaluate
and hookRunParallel
have changed.
See help(hook.run.default)
and help(hook.evaluate.default)
for examples of
the new interfaces.
Printing of race progress now reports the actual configuration and instance IDs, and numbers are printed in a more human-readable format. (Leslie Pérez Cáceres, Manuel López-Ibáñez)
Reduce memory use for very large values of maxExperiments
.
(Manuel López-Ibáñez, thanks to Federico Caselli for identifying the issue)
New option --load-balancing
(loadBalancing
) for disabling load-balancing
when executing jobs in parallel. Load-balancing makes better use of
computing resources, but increases communication overhead. If this overhead
is large, disabling load-balancing may be faster.
(Manuel López-Ibáñez, thanks to Federico Caselli for identifying the issue)
The option --parallel
in Windows now uses load-balancing by default.
(Manuel López-Ibáñez)
The wall-clock time after finishing each task is printed in the output. (Manuel López-Ibáñez, thanks to Federico Caselli for providing an initial patch)
Fix bug that could introduce spurious whitespace when printing the final configurations. (Manuel López-Ibáñez)
Fix bug if there are more initial candidates than needed for the first race. (Leslie Pérez Cáceres, Manuel López-Ibáñez)
New configuration options, mainly for R users:
hookRunParallel
: Optional R function to provide custom
parallelization of hook.run
.
hookRunData
: Optional data passed to hookRun
. This is ignored by the
default hookRun
function, but it may be used by custom hookRun
R
functions to pass persistent data around. (Manuel López-Ibáñez)
New option --version
. (Manuel López-Ibáñez)
Terminate early if there is no sufficient budget to run irace with the given settings. (Manuel López-Ibáñez)
The option --parallel
(without --mpi
) now works under Windows.
(Manuel López-Ibáñez, thanks to Pablo Valledor Pellicer for testing
it)
Improved error handling when running under Rmpi. Now irace will terminate as soon as the master node detects at least one failed slave node. This avoids irace reporting two times the same error. Also, irace will print all the unique errors returned by all slaves and not just the first one. (Manuel López-Ibáñez)
Forbidden configurations may be specified in terms of constraints
on their values. Forbidden configurations will never be evaluated by irace.
See --forbidden-file
and inst/templates/forbidden.tmpl
.
(Manuel López-Ibáñez)
New option --recovery-file
(recoveryFile
) allows resuming a
previous irace run. (Leslie Pérez Cáceres)
The confidence level for the elimination test is now
configurable with parameter --confidence
. (Leslie Pérez Cáceres)
Much more robust handling of relative/absolute paths. Improved support for Windows. (Leslie Pérez Cáceres, Manuel López-Ibáñez)
Provide better error messages for incorrect parameter descriptions. (Manuel López-Ibáñez) Examples:
x "" i (0, 0) # lower and upper bounds are the same
x "" r (1e-4, 5e-4) # given digits=2, ditto
x "" i (-1, -2) # lower bound must be smaller than upper bound
x "" c ("a", "a") # duplicated values
Print elapsed time for calls to hook-run if debugLevel >=1
.
(Manuel López-Ibáñez)
examples/hook-run-python/hook-run
: A multi-purpose hook-run
written
in Python. (Franco Mascia)
Parallel mode in an SGE cluster (--sge-cluster
) is more
robust. (Manuel López-Ibáñez)
Replace obsolete package multicore by package parallel (requires R >= 2.14.0)
Use load-balancing (mc.preschedule = FALSE
) in mclapply
.
Use reg.finalizer
to finish Rmpi properly without clobbering
.Last()
.
Remove uses of deprecated as.real()
.
Nicer error handling in readParameters()
.
Add hypervolume (multi-objective) example.
Fix several bugs in the computation of similar candidates.
More concise output.
The parameters expName
and expDescription
are now useless and they
were removed.
Faster computation of similar candidates (Jeremie Dubois-Lacoste and Leslie Pérez Cáceres).
Fix bug when saving instances in tunerResults$experiments
.
irace.cmdline ("--help")
does not try to quit R anymore.
Fix bug caused by file.exists (and possibly other functions) not handling directory names with a trailing backslash or slash on Windows.
Fix bug using per-instance parameters (Leslie Pérez Cáceres).
Fix bug when reading initial candidates from a file.
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.