nojhan/paradiseo
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

feat: add tailored irace config files

Browse source
This commit is contained in:
Johann Dreo 2020-10-06 15:44:08 +02:00
commit 2c80a01fad
4 changed files with 378 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

48
eo/contrib/irace/irace-config/default.instances Normal file
View file

@ -0,0 +1,48 @@
## This is an example of specifying instances with a file.
# Each line is an instance relative to trainInstancesDir
# (see scenario.txt.tmpl) and an optional sequence of instance-specific
# parameters that will be passed to target-runnerx when invoked on that
# instance.
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

227
eo/contrib/irace/irace-config/example.scen Normal file
View file

@ -0,0 +1,227 @@
###################################################### -*- mode: r -*- #####
## Scenario setup for Iterated Race (irace).
############################################################################
## To use the default value of a parameter of iRace, simply do not set
## the parameter (comment it out in this file, and do not give any
## value on the command line).
## File that contains the description of the parameters of the target
## algorithm.
parameterFile = "./fastga.param"
## Directory where the programs will be run.
execDir = "."
## File to save tuning results as an R dataset, either absolute path or
## relative to execDir.
# logFile = "./irace.Rdata"
## Previously saved log file to recover the execution of irace, either
## absolute path or relative to the current directory. If empty or NULL,
## recovery is not performed.
# recoveryFile = ""
## Directory where training instances are located; either absolute path or
## relative to current directory. If no trainInstancesFiles is provided,
## all the files in trainInstancesDir will be listed as instances.
trainInstancesDir = "."
## File that contains a list of training instances and optionally
## additional parameters for them. If trainInstancesDir is provided, irace
## will search for the files in this folder.
trainInstancesFile = "./default.instances"
## File that contains a table of initial configurations. If empty or NULL,
## all initial configurations are randomly generated.
# configurationsFile = ""
## File that contains a list of logical expressions that cannot be TRUE
## for any evaluated configuration. If empty or NULL, do not use forbidden
## expressions.
# forbiddenFile = ""
## Script called for each configuration that executes the target algorithm
## to be tuned. See templates.
targetRunner = "./target-runner"
## Number of times to retry a call to targetRunner if the call failed.
# targetRunnerRetries = 0
## Optional data passed to targetRunner. This is ignored by the default
## targetRunner function, but it may be used by custom targetRunner
## functions to pass persistent data around.
# targetRunnerData = ""
## Optional R function to provide custom parallelization of targetRunner.
# targetRunnerParallel = ""
## Optional script or R function that provides a numeric value for each
## configuration. See templates/target-evaluator.tmpl
# targetEvaluator = ""
## Maximum number of runs (invocations of targetRunner) that will be
## performed. It determines the maximum budget of experiments for the
## tuning.
maxExperiments = 2000
## Maximum total execution time in seconds for the executions of
## targetRunner. targetRunner must return two values: cost and time.
# maxTime = 60
## Fraction (smaller than 1) of the budget used to estimate the mean
## computation time of a configuration. Only used when maxTime > 0
# budgetEstimation = 0.02
## Maximum number of decimal places that are significant for numerical
## (real) parameters.
digits = 2
## Debug level of the output of irace. Set this to 0 to silence all debug
## messages. Higher values provide more verbose debug messages.
# debugLevel = 0
## Number of iterations.
# nbIterations = 0
## Number of runs of the target algorithm per iteration.
# nbExperimentsPerIteration = 0
## Randomly sample the training instances or use them in the order given.
# sampleInstances = 1
## Statistical test used for elimination. Default test is always F-test
## unless capping is enabled, in which case the default test is t-test.
## Valid values are: F-test (Friedman test), t-test (pairwise t-tests with
## no correction), t-test-bonferroni (t-test with Bonferroni's correction
## for multiple comparisons), t-test-holm (t-test with Holm's correction
## for multiple comparisons).
# testType = "F-test"
## Number of instances evaluated before the first elimination test. It
## must be a multiple of eachTest.
# firstTest = 5
## Number of instances evaluated between elimination tests.
# eachTest = 1
## Minimum number of configurations needed to continue the execution of
## each race (iteration).
# minNbSurvival = 0
## Number of configurations to be sampled and evaluated at each iteration.
# nbConfigurations = 0
## Parameter used to define the number of configurations sampled and
## evaluated at each iteration.
# mu = 5
## Confidence level for the elimination test.
# confidence = 0.95
## If the target algorithm is deterministic, configurations will be
## evaluated only once per instance.
# deterministic = 0
## Seed of the random number generator (by default, generate a random
## seed).
# seed = NA
## Number of calls to targetRunner to execute in parallel. Values 0 or 1
## mean no parallelization.
# parallel = 0
## Enable/disable load-balancing when executing experiments 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.
# loadBalancing = 1
## Enable/disable MPI. Use Rmpi to execute targetRunner in parallel
## (parameter parallel is the number of slaves).
# mpi = 0
## Specify how irace waits for jobs to finish when targetRunner submits
## jobs to a batch cluster: sge, pbs, torque or slurm. targetRunner must
## submit jobs to the cluster using, for example, qsub.
# batchmode = 0
## Enable/disable the soft restart strategy that avoids premature
## convergence of the probabilistic model.
# softRestart = 1
## Soft restart threshold value for numerical parameters. If NA, NULL or
## "", it is computed as 10^-digits.
# softRestartThreshold = ""
## Directory where testing instances are located, either absolute or
## relative to current directory.
# testInstancesDir = ""
## File containing a list of test instances and optionally additional
## parameters for them.
# testInstancesFile = ""
## Number of elite configurations returned by irace that will be tested if
## test instances are provided.
# testNbElites = 1
## Enable/disable testing the elite configurations found at each
## iteration.
# testIterationElites = 0
## Enable/disable elitist irace.
# elitist = 1
## Number of instances added to the execution list before previous
## instances in elitist irace.
# elitistNewInstances = 1
## In elitist irace, maximum number per race of elimination tests that do
## not eliminate a configuration. Use 0 for no limit.
# elitistLimit = 2
## User-defined R function that takes a configuration generated by irace
## and repairs it.
# repairConfiguration = ""
## Enable the use of adaptive capping, a technique designed for minimizing
## the computation time of configurations. This is only available when
## elitist is active.
# capping = 0
## Measure used to obtain the execution bound from the performance of the
## elite configurations: median, mean, worst, best.
# cappingType = "median"
## Method to calculate the mean performance of elite configurations:
## candidate or instance.
# boundType = "candidate"
## Maximum execution bound for targetRunner. It must be specified when
## capping is enabled.
# boundMax = 0
## Precision used for calculating the execution time. It must be specified
## when capping is enabled.
# boundDigits = 0
## Penalization constant for timed out executions (executions that reach
## boundMax execution time).
# boundPar = 1
## Replace the configuration cost of bounded executions with boundMax.
# boundAsTimeout = 1
## Percentage of the configuration budget used to perform a postselection
## race of the best configurations of each iteration after the execution
## of irace.
# postselection = 0
## Enable/disable AClib mode. This option enables compatibility with
## GenericWrapper4AC as targetRunner script.
# aclib = 0
## END of scenario file
############################################################################

84
eo/contrib/irace/irace-config/target-runner Executable file
View file

@ -0,0 +1,84 @@
#!/bin/bash
###############################################################################
# This script is the command that is executed every run.
# Check the examples in examples/
#
# This script is run in the execution directory (execDir, --exec-dir).
#
# PARAMETERS:
# $1 is the candidate configuration number
# $2 is the instance ID
# $3 is the seed
# $4 is the instance name
# The rest ($* after `shift 4') are parameters to the run
#
# RETURN VALUE:
# This script should print one numerical value: the cost that must be minimized.
# Exit with 0 if no error, with 1 in case of error
###############################################################################
error() {
echo "`TZ=UTC date`: $0: error: $@"
exit 1
}
EXE="./fastga"
FIXED_PARAMS="--problem=0"
CONFIG_ID=$1
INSTANCE_ID=$2
SEED=$3
INSTANCE=$(echo $4 | sed 's/\//\n/g'|tail -n 1)
CROSSOVER_RATE=$5
CROSSOVER_SELECTOR=$6
CROSSOVER=$7
MUTATION_RATE=$8
MUT_SELECTOR=$9
MUTATION=${10}
REPLACEMENT=${11}
shift 11 || error "Not enough parameters"
INSTANCE_PARAMS=$*
STDOUT=c${CONFIG_ID}_i${INSTANCE_ID}_s${SEED}.stdout
STDERR=c${CONFIG_ID}_i${INSTANCE_ID}_s${SEED}.stderr
if [ ! -x "${EXE}" ]; then
error "${EXE}: not found or not executable (pwd: $(pwd))"
fi
# If the program just prints a number, we can use 'exec' to avoid
# creating another process, but there can be no other commands after exec.
#exec $EXE ${FIXED_PARAMS} -i $INSTANCE ${INSTANCE_PARAMS}
# exit 1
#
# Otherwise, save the output to a file, and parse the result from it.
# (If you wish to ignore segmentation faults you can use '{}' around
# the command.)
cmd="$EXE ${FIXED_PARAMS} --instance=${INSTANCE} --seed=${SEED} ${CROSSOVER_RATE} ${CROSSOVER_SELECTOR} ${CROSSOVER} ${MUTATION_RATE} ${MUT_SELECTOR} ${MUTATION} ${REPLACEMENT}"
# NOTE: irace seems to capture both stderr and stdout, so you should not output to stderr
echo ${cmd} > ${STDERR}
$cmd 2> ${STDERR} | tee ${STDOUT}
# The following code is useless if the binary only output a single number on stdout.
# This may be used to introduce a delay if there are filesystem
# issues.
# SLEEPTIME=1
# while [ ! -s "${STDOUT}" ]; do
# sleep $SLEEPTIME
# let "SLEEPTIME += 1"
# done
# This is an example of reading a number from the output.
# It assumes that the objective value is the first number in
# the first column of the last line of the output.
# if [ -s "${STDOUT}" ]; then
# COST=$(tail -n 1 ${STDOUT} | grep -e '^[[:space:]]*[+-]\?[0-9]' | cut -f1)
# echo "$COST"
# rm -f "${STDOUT}" "${STDERR}"
# exit 0
# else
# error "${STDOUT}: No such file or directory"
# fi

19
eo/contrib/irace/run_irace.sh Executable file
View file

@ -0,0 +1,19 @@
#!/bin/bash
if [[ $# != 1 ]] ; then
echo "ERROR: build dir not indicated"
exit 1
fi
cd $1
pwd
# Fore some reason, irace absolutely need those files...
cp ../irace-config/example.scen .
cp ../irace-config/target-runner .
cp ../irace-config/default.instances .
# Generate the parameter list file.
./fastga -h > fastga.param 2>/dev/null
/usr/lib/R/site-library/irace/bin/irace --scenario example.scen