def tableonealg(dsList, allmintarget, allertbest, sortedAlgs=None,
outputdir='.'):
"""Routine for the generation of a table for an algorithm."""
header2 = ('evals/D', '\%trials', '\%inst', '\multicolumn{2}{c|}{fcts}', 'best', '10', '25', 'med', '75', '90')
format2 = ('%.3g', '%d', '%d', '%d', '%d', '%1.1e', '%1.1e', '%1.1e', '%1.1e', '%1.1e', '%1.1e')
ilines = [r'\begin{tabular}{cccc@{/}c|cccccc}',
r'\multicolumn{5}{c|}{Solved} & \multicolumn{6}{|c}{ERT/ERT$_{\textrm{best}}$} \\',
' & '.join(header2)]
dictDim = dsList.dictByDim()
for d, dentries in dictDim.iteritems():
dictAlg = dentries.dictByAlg()
# one-alg table
for alg in sortedAlgs:
# Regroup entries by algorithm
lines = ilines[:]
algentries = DataSetList()
for i in alg:
if dictAlg.has_key(i):
algentries.extend(dictAlg[i])
table = onealg(algentries, allmintarget, allertbest)
for i in table:
lines[-1] += r'\\'
if numpy.isinf(i[0]):
tmpstr = r'$\infty$'
else:
tmpstr = format2[0] % i[0]
for j in range(1, 5):
tmpstr += (' & %s' % format2[j]) % i[j]
for j in range(5, len(i)):
tmpstr += ' & %s' % writeFEvals(i[j])
lines.append(tmpstr)
lines.append(r'\end{tabular}')
f = open(os.path.join(outputdir, 'pptable_%s_%02dD.tex' % (algShortInfos[alg[0]], d)), 'w')
# any element of alg would convene.
f.write('\n'.join(lines) + '\n')
f.close()
def tablemanyalg(dsList, allmintarget, allertbest, sortedAlgs=None,
outputdir='.'):
"""Generate a table with the figures of multiple algorithms."""
stargets = sorted(allmintarget.keys())
dictDim = dsList.dictByDim()
maxRank = 3
for d, dentries in dictDim.iteritems():
dictAlg = dentries.dictByAlg()
# Multiple algorithms table.
# Generate data
table = []
algnames = []
for alg in sortedAlgs:
# Regroup entries by algorithm
algentries = DataSetList()
for i in alg:
if dictAlg.has_key(i):
algentries.extend(dictAlg[i])
if not algentries:
continue
algnames.append(writeLabels(algPlotInfos[alg[0]]['label']))
tmp = []
for t in stargets:
dictFunc = algentries.dictByFunc()
erts = []
for func, entry in dictFunc.iteritems():
try:
entry = entry[0]
except:
raise Usage('oops too many entries')
try:
if numpy.isnan(allmintarget[t][(func, d)]):
continue
except LookupError:
continue
# At this point the target exists.
try:
erts.append(entry.ert[entry.target<=allmintarget[t][(func, d)]][0]/allertbest[t][(func, d)])
except LookupError:
erts.append(numpy.inf)
if numpy.isfinite(erts).any():
tmp += [numpy.median(erts), numpy.min(erts), numpy.sum(numpy.isfinite(erts))]
else:
tmp += [numpy.inf, numpy.inf, 0]
table.append(tmp)
# Process over all data
table = numpy.array(table)
kept = [] # range(numpy.shape(table)[1])
targetkept = []
for i in range(1, (numpy.shape(table)[1])/3 + 1):
if (table[:, 3*i - 1] != 0).any():
kept.extend([3*i - 3, 3*i - 2 , 3*i - 1])
targetkept.append(i-1)
table = table[:, kept]
#set_trace()
dtype = []
for i, t in enumerate(stargets):
dtype.extend([('med%d' % i, 'f4'), ('min%d' % i, 'f4'),
('nbsolved%d' % i, 'i1')])
dtype = list(dtype[i] for i in kept)
boldface = sortColumns(table, maxRank)
idxsort = numpy.argsort(numpy.array(list(tuple(i) for i in table),
dtype=dtype),
order=('med4', 'med2', 'med0', 'min0'))
# Sorted successively by med(ERT) / ERTbest for fevals/D = 100, 10, 1
# and then min(ERT) / ERTbest for fevals/D = 1
# format the data
lines = [r'\begin{tabular}{c' + 'c@{/}c@{(}c@{) }'*len(targetkept) + '}']
tmpstr = 'evals/D'
for t in list(stargets[i] for i in targetkept):
nbsolved = sum(numpy.isfinite(list(allmintarget[t][i] for i in allmintarget[t] if i[1] == d)))
#set_trace()
tmpstr += (r' & \multicolumn{2}{c@{(}}{%s} & %d' % (writeFEvals(t), nbsolved))
lines.append(tmpstr)
for i in idxsort:
line = table[i]
lines[-1] += r'\\'
curline = algnames[i]
for j in range(len(table[i])):
curline += ' & '
if (j + 1) % 3 > 0: # the test may not be necessary
if numpy.isinf(line[j]):
tmpstr = '.'
else:
tmpstr = '%s' % (writeFEvals(line[j]))
if i in boldface[j]:
tmpstr = r'\textbf{' + tmpstr + '}'
curline += tmpstr
else:
curline += '%d' % line[j] # nb solved.
lines.append(curline)
lines.append(r'\end{tabular}')
f = open(os.path.join(outputdir, 'pptableall_%02dD.tex' % (d)), 'w')
f.write('\n'.join(lines) + '\n')
f.close()
lines.append('\\providecommand{\\algorithm' + abc[i] + '}{' +
str_to_latex(strip_pathname2(alg)) + '}')
prepend_to_file(
os.path.join(outputdir, 'bbob_pproc_commands.tex'), lines, 5000,
'bbob_proc_commands.tex truncated, consider removing the file before the text run'
)
dsList, sortedAlgs, dictAlg = processInputArgs(args, verbose=verbose)
if not dsList:
sys.exit()
for i in dictAlg:
if isNoisy and not isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'nzall', DataSetList())
if isNoiseFree and not isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'noiselessall', DataSetList())
for i in dsList:
if i.dim not in genericsettings.dimensions_to_display:
continue
if (dict((j, i.instancenumbers.count(j))
for j in set(i.instancenumbers)) <
inset.instancesOfInterest):
warnings.warn('The data of %s do not list ' % (i) +
'the correct instances ' + 'of function F%d.' %
(i.funcId))
def main(argv=None):
r"""Routine for post-processing COCO data from two algorithms.
Provided with some data, this routine outputs figure and TeX files
in a folder needed for the compilation of latex document
:file:`template2XXX.tex` or :file:`noisytemplate2XXX.tex`, where
:file:`XXX` is either :file:`ecj` or :file:`generic`. The template
file needs to be edited so that the command ``\bbobdatapath`` points
to the output folder.
These output files will contain performance tables, performance
scaling figures and empirical cumulative distribution figures. On
subsequent executions, new files will be added to the output folder,
overwriting existing older files in the process.
Keyword arguments:
*argv* -- list of strings containing options and arguments. If not
given, sys.argv is accessed.
*argv* must list folders containing BBOB data files. Each of these
folders should correspond to the data of ONE algorithm.
Furthermore, argv can begin with, in any order, facultative option
flags listed below.
-h, --help
displays this message.
-v, --verbose
verbose mode, prints out operations.
-o OUTPUTDIR, --output-dir=OUTPUTDIR
changes the default output directory (:file:`ppdata`) to
:file:`OUTPUTDIR`
--noise-free, --noisy
processes only part of the data.
--settings=SETTING
changes the style of the output figures and tables. At the
moment only the only differences are in the colors of the
output figures. SETTING can be either "grayscale", "color"
or "black-white". The default setting is "color".
--fig-only, --rld-only, --tab-only, --sca-only
these options can be used to output respectively the ERT
graphs figures, run length distribution figures or the
comparison tables scatter plot figures only. Any combination
of these options results in no output.
--conv
if this option is chosen, additionally convergence
plots for each function and algorithm are generated.
Exceptions raised:
*Usage* -- Gives back a usage message.
Examples:
* Calling the rungeneric2.py interface from the command line::
$ python bbob_pproc/rungeneric2.py -v Alg0-baseline Alg1-of-interest
will post-process the data from folders :file:`Alg0-baseline` and
:file:`Alg1-of-interest`, the former containing data for the
reference algorithm (zero-th) and the latter data for the
algorithm of concern (first). The results will be output in the
default output folder. The ``-v`` option adds verbosity.
* From the python interpreter (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.rungeneric2.main('-o outputfolder PSO DEPSO'.split())
This will execute the post-processing on the data found in folder
:file:`PSO` and :file:`DEPSO`. The ``-o`` option changes the output
folder from the default to :file:`outputfolder`.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
try:
try:
opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
isfigure = True
isrldistr = True
istable = True
isscatter = True
isscaleup = True
isNoisy = False
isNoiseFree = False
verbose = False
outputdir = 'ppdata'
inputsettings = 'color'
isConv= False
#Process options
for o, a in opts:
if o in ("-v","--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-o", "--output-dir"):
outputdir = a
#elif o in ("-s", "--style"):
# inputsettings = a
elif o == "--fig-only":
isrldistr = False
istable = False
isscatter = False
elif o == "--rld-only":
isfigure = False
istable = False
isscatter = False
elif o == "--tab-only":
isfigure = False
isrldistr = False
isscatter = False
elif o == "--sca-only":
isfigure = False
isrldistr = False
istable = False
elif o == "--noisy":
isNoisy = True
elif o == "--noise-free":
isNoiseFree = True
elif o == "--settings":
inputsettings = a
elif o == "--conv":
isConv = True
else:
assert False, "unhandled option"
# from bbob_pproc import bbob2010 as inset # input settings
if inputsettings == "color":
from bbob_pproc import config, genericsettings as inset # input settings
config.config()
elif inputsettings == "grayscale":
from bbob_pproc import grayscalesettings as inset # input settings
elif inputsettings == "black-white":
from bbob_pproc import bwsettings as inset # input settings
else:
txt = ('Settings: %s is not an appropriate ' % inputsettings
+ 'argument for input flag "--settings".')
raise Usage(txt)
if (not verbose):
warnings.simplefilter('module')
warnings.simplefilter('ignore')
print ("Post-processing will generate comparison " +
"data in folder %s" % outputdir)
print " this might take several minutes."
dsList, sortedAlgs, dictAlg = processInputArgs(args, verbose=verbose)
if 1 < 3 and len(sortedAlgs) != 2:
raise ValueError('rungeneric2.py needs exactly two algorithms to compare, found: '
+ str(sortedAlgs)
+ '\n use rungeneric.py (or rungenericmany.py) to compare more algorithms. ')
if not dsList:
sys.exit()
for i in dictAlg:
if isNoisy and not isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get('nzall', DataSetList())
if isNoiseFree and not isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get('noiselessall', DataSetList())
for i in dsList:
if i.dim not in genericsettings.dimensions_to_display:
continue
if (dict((j, i.instancenumbers.count(j)) for j in set(i.instancenumbers)) <
inset.instancesOfInterest):
warnings.warn('The data of %s do not list ' %(i) +
'the correct instances ' +
'of function F%d.' %(i.funcId))
if len(sortedAlgs) < 2:
raise Usage('Expect data from two different algorithms, could ' +
'only find one.')
elif len(sortedAlgs) > 2:
warnings.warn('Data from folders: %s ' % (sortedAlgs) +
'were found, the first two will be processed.')
# Group by algorithm
dsList0 = dictAlg[sortedAlgs[0]]
if not dsList0:
raise Usage('Could not find data for algorithm %s.' % (sortedAlgs[0]))
dsList1 = dictAlg[sortedAlgs[1]]
if not dsList1:
raise Usage('Could not find data for algorithm %s.' % (sortedAlgs[0]))
# get the name of each algorithm from the input arguments
tmppath0, alg0name = os.path.split(sortedAlgs[0].rstrip(os.sep))
tmppath1, alg1name = os.path.split(sortedAlgs[1].rstrip(os.sep))
for i in dsList0:
i.algId = alg0name
for i in dsList1:
i.algId = alg1name
######################### Post-processing #############################
if isfigure or isrldistr or istable or isscatter:
if not os.path.exists(outputdir):
os.mkdir(outputdir)
if verbose:
print 'Folder %s was created.' % (outputdir)
# prepend the algorithm name command to the tex-command file
abc = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
lines = []
for i, alg in enumerate(args):
lines.append('\\providecommand{\\algorithm' + abc[i] + '}{' +
str_to_latex(strip_pathname(alg)) + '}')
prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'),
lines, 1000,
'bbob_proc_commands.tex truncated, consider removing the file before the text run'
)
# Check whether both input arguments list noisy and noise-free data
dictFN0 = dsList0.dictByNoise()
dictFN1 = dsList1.dictByNoise()
k0 = set(dictFN0.keys())
k1 = set(dictFN1.keys())
symdiff = k1 ^ k0 # symmetric difference
if symdiff:
tmpdict = {}
for i, noisegrp in enumerate(symdiff):
if noisegrp == 'nzall':
tmp = 'noisy'
elif noisegrp == 'noiselessall':
tmp = 'noiseless'
if dictFN0.has_key(noisegrp):
tmp2 = sortedAlgs[0]
elif dictFN1.has_key(noisegrp):
tmp2 = sortedAlgs[1]
tmpdict.setdefault(tmp2, []).append(tmp)
txt = []
for i, j in tmpdict.iteritems():
txt.append('Only input folder %s lists %s data.'
% (i, ' and '.join(j)))
raise Usage('Data Mismatch: \n ' + ' '.join(txt)
+ '\nTry using --noise-free or --noisy flags.')
if isfigure:
plt.rc("axes", **inset.rcaxeslarger)
plt.rc("xtick", **inset.rcticklarger)
plt.rc("ytick", **inset.rcticklarger)
plt.rc("font", **inset.rcfontlarger)
plt.rc("legend", **inset.rclegendlarger)
ppfig2.main(dsList0, dsList1, ftarget, outputdir, verbose)
print "log ERT1/ERT0 vs target function values done."
plt.rc("axes", **inset.rcaxes)
plt.rc("xtick", **inset.rctick)
plt.rc("ytick", **inset.rctick)
plt.rc("font", **inset.rcfont)
plt.rc("legend", **inset.rclegend)
if isrldistr:
if len(dictFN0) > 1 or len(dictFN1) > 1:
warnings.warn('Data for functions from both the noisy and ' +
'non-noisy testbeds have been found. Their ' +
'results will be mixed in the "all functions" ' +
'ECDF figures.')
dictDim0 = dsList0.dictByDim()
dictDim1 = dsList1.dictByDim()
# ECDFs of ERT ratios
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
if dim in inset.rldDimsOfInterest:
# ECDF for all functions altogether
try:
pprldistr2.main(dictDim0[dim], dictDim1[dim],
inset.rldValsOfInterest,
outputdir, '%02dD_all' % dim, verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.'
% (dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr2.main(dictFG1[fGroup], dictFG0[fGroup],
inset.rldValsOfInterest,
outputdir, '%02dD_%s' % (dim, fGroup),
verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr2.main(dictFN1[fGroup], dictFN0[fGroup],
inset.rldValsOfInterest, outputdir,
'%02dD_%s' % (dim, fGroup),
verbose)
print "ECDF runlength ratio graphs done."
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
pprldistr.fmax = None #Resetting the max final value
pprldistr.evalfmax = None #Resetting the max #fevalsfactor
# ECDFs of all functions altogether
if dim in inset.rldDimsOfInterest:
try:
pprldistr.comp(dictDim1[dim], dictDim0[dim],
inset.rldValsOfInterest if isinstance(inset.rldValsOfInterest, TargetValues)
else TargetValues(inset.rldValsOfInterest),
True,
outputdir, 'all', verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.'
% (dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr.comp(dictFG1[fGroup], dictFG0[fGroup],
inset.rldValsOfInterest if isinstance(inset.rldValsOfInterest, TargetValues)
else TargetValues(inset.rldValsOfInterest),
True, outputdir,
'%s' % fGroup, verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr.comp(dictFN1[fGroup], dictFN0[fGroup],
inset.rldValsOfInterest if isinstance(inset.rldValsOfInterest, TargetValues)
else TargetValues(inset.rldValsOfInterest),
True, outputdir,
'%s' % fGroup, verbose)
print "ECDF runlength graphs done."
if isConv:
ppconverrorbars.main(dictAlg,outputdir,verbose)
if istable:
dictNG0 = dsList0.dictByNoise()
dictNG1 = dsList1.dictByNoise()
for nGroup in set(dictNG0.keys()) & set(dictNG1.keys()):
# split table in as many as necessary
dictFunc0 = dictNG0[nGroup].dictByFunc()
dictFunc1 = dictNG1[nGroup].dictByFunc()
funcs = list(set(dictFunc0.keys()) & set(dictFunc1.keys()))
if len(funcs) > 24:
funcs.sort()
nbgroups = int(numpy.ceil(len(funcs)/24.))
def split_seq(seq, nbgroups):
newseq = []
splitsize = 1.0/nbgroups*len(seq)
for i in range(nbgroups):
newseq.append(seq[int(round(i*splitsize)):int(round((i+1)*splitsize))])
return newseq
groups = split_seq(funcs, nbgroups)
# merge
group0 = []
group1 = []
for i, g in enumerate(groups):
tmp0 = DataSetList()
tmp1 = DataSetList()
for f in g:
tmp0.extend(dictFunc0[f])
tmp1.extend(dictFunc1[f])
group0.append(tmp0)
group1.append(tmp1)
for i, g in enumerate(zip(group0, group1)):
pptable2.main(g[0], g[1], inset.tabDimsOfInterest,
outputdir, '%s%d' % (nGroup, i), verbose)
else:
if 11 < 3: # future handling:
dictFunc0 = dsList0.dictByFunc()
dictFunc1 = dsList1.dictByFunc()
funcs = list(set(dictFunc0.keys()) & set(dictFunc1.keys()))
funcs.sort()
# nbgroups = int(numpy.ceil(len(funcs)/testbedsettings.numberOfFunctions))
# pptable2.main(dsList0, dsList1,
# testbedsettings.tabDimsOfInterest, outputdir,
# '%s' % (testbedsettings.testbedshortname), verbose)
else:
pptable2.main(dictNG0[nGroup], dictNG1[nGroup],
inset.tabDimsOfInterest, outputdir,
'%s' % (nGroup), verbose)
prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'),
['\\providecommand{\\bbobpptablestwolegend}[1]{',
pptable2.figure_legend,
'}'
])
print "Tables done."
if isscatter:
ppscatter.main(dsList0, dsList1, outputdir, verbose=verbose)
prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'),
['\\providecommand{\\bbobppscatterlegend}[1]{',
ppscatter.figure_legend,
'}'
])
print "Scatter plots done."
if isscaleup:
plt.rc("axes", labelsize=20, titlesize=24)
plt.rc("xtick", labelsize=20)
plt.rc("ytick", labelsize=20)
plt.rc("font", size=20)
plt.rc("legend", fontsize=20)
ppfigs.main(dictAlg, sortedAlgs, ftarget, outputdir, verbose)
plt.rcdefaults()
print "Scaling figures done."
if isfigure or isrldistr or istable or isscatter or isscaleup:
print "Output data written to folder %s" % outputdir
plt.rcdefaults()
def main(argv=None):
r"""Post-processing COCO data of a single algorithm.
Provided with some data, this routine outputs figure and TeX files
in a folder needed for the compilation of latex document
:file:`template1XXX.tex` or :file:`noisytemplate1XXX.tex`, where
:file:`XXX` is either :file:`ecj` or :file:`generic`. The template
file needs to be edited so that the commands
``\bbobdatapath`` and ``\algfolder`` point to the output folder.
These output files will contain performance tables, performance
scaling figures and empirical cumulative distribution figures. On
subsequent executions, new files will be added to the output folder,
overwriting existing older files in the process.
Keyword arguments:
*argv* -- list of strings containing options and arguments. If not
given, sys.argv is accessed.
*argv* should list either names of :file:`info` files or folders
containing :file:`info` files. argv can also contain post-processed
:file:`pickle` files generated by this routine. Furthermore, *argv*
can begin with, in any order, facultative option flags listed below.
-h, --help
displays this message.
-v, --verbose
verbose mode, prints out all operations.
-p, --pickle
generates pickle post processed data files.
-o OUTPUTDIR, --output-dir=OUTPUTDIR
changes the default output directory (:file:`ppdata`) to
:file:`OUTPUTDIR`.
--crafting-effort=VALUE
sets the crafting effort to VALUE (float). Otherwise the
default value of 0. will be used.
--noise-free, --noisy
processes only part of the data.
--settings=SETTINGS
changes the style of the output figures and tables. At the
moment the only differences are in the colors of the output
figures. SETTINGS can be either "grayscale", "color" or
"black-white". The default setting is "color".
--tab-only, --fig-only, --rld-only, --los-only
these options can be used to output respectively the TeX
tables, convergence and ERTs graphs figures, run length
distribution figures, ERT loss ratio figures only. A
combination of any two of these options results in no
output.
--conv
if this option is chosen, additionally convergence plots
for each function and algorithm are generated.
--expensive
runlength-based f-target values and fixed display limits,
useful with comparatively small budgets. By default the
setting is based on the budget used in the data.
--not-expensive
expensive setting off.
--runlength-based
runlength-based f-target values, such that the
"level of difficulty" is similar for all functions.
Exceptions raised:
*Usage* -- Gives back a usage message.
Examples:
* Calling the rungeneric1.py interface from the command line::
$ python bbob_pproc/rungeneric1.py -v experiment1
will post-process the folder experiment1 and all its containing
data, base on the .info files found in the folder. The result will
appear in the default output folder. The -v option adds verbosity. ::
$ python bbob_pproc/rungeneric1.py -o exp2 experiment2/*.info
This will execute the post-processing on the info files found in
:file:`experiment2`. The result will be located in the alternative
location :file:`exp2`.
* Loading this package and calling the main from the command line
(requires that the path to this package is in python search path)::
$ python -m bbob_pproc.rungeneric1 -h
This will print out this help message.
* From the python interpreter (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.rungeneric1.main('-o outputfolder folder1'.split())
This will execute the post-processing on the index files found in
:file:`folder1`. The ``-o`` option changes the output folder from
the default to :file:`outputfolder`.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
if 1 < 3:
opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
if 11 < 3:
try:
opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
except getopt.error, msg:
raise Usage(msg)
if not (args) and not '--help' in argv and not 'h' in argv:
print 'not enough input arguments given'
print 'cave: the following options also need an argument:'
print[o for o in longoptlist if o[-1] == '=']
print 'options given:'
print opts
print 'try --help for help'
sys.exit()
inputCrE = 0.
isfigure = True
istab = True
isrldistr = True
islogloss = True
isPostProcessed = False
isPickled = False
verbose = False
outputdir = 'ppdata'
isNoisy = False
isNoiseFree = False
inputsettings = 'color'
isConv = False
isRLbased = None # allows automatic choice
isExpensive = None
# Process options
for o, a in opts:
if o in ("-v", "--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-p", "--pickle"):
isPickled = True
elif o in ("-o", "--output-dir"):
outputdir = a
elif o == "--noisy":
isNoisy = True
elif o == "--noise-free":
isNoiseFree = True
# The next 4 are for testing purpose
elif o == "--tab-only":
isfigure = False
isrldistr = False
islogloss = False
elif o == "--fig-only":
istab = False
isrldistr = False
islogloss = False
elif o == "--rld-only":
istab = False
isfigure = False
islogloss = False
elif o == "--los-only":
istab = False
isfigure = False
isrldistr = False
elif o == "--crafting-effort":
try:
inputCrE = float(a)
except ValueError:
raise Usage(
'Expect a valid float for flag crafting-effort.')
elif o == "--settings":
inputsettings = a
elif o == "--conv":
isConv = True
elif o == "--runlength-based":
isRLbased = True
elif o == "--expensive":
isExpensive = True # comprises runlength-based
elif o == "--not-expensive":
isExpensive = False
else:
assert False, "unhandled option"
# from bbob_pproc import bbob2010 as inset # input settings
if inputsettings == "color":
from bbob_pproc import genericsettings as inset # input settings
elif inputsettings == "grayscale":
from bbob_pproc import grayscalesettings as inset # input settings
elif inputsettings == "black-white":
from bbob_pproc import bwsettings as inset # input settings
else:
txt = ('Settings: %s is not an appropriate ' % inputsettings +
'argument for input flag "--settings".')
raise Usage(txt)
if 11 < 3:
from bbob_pproc import config # input settings
config.config()
import imp
# import testbedsettings as testbedsettings # input settings
try:
fp, pathname, description = imp.find_module("testbedsettings")
testbedsettings = imp.load_module("testbedsettings", fp,
pathname, description)
finally:
fp.close()
if (not verbose):
warnings.simplefilter('module')
# warnings.simplefilter('ignore')
print("Post-processing (1): will generate output " +
"data in folder %s" % outputdir)
print " this might take several minutes."
filelist = list()
for i in args:
i = i.strip()
if os.path.isdir(i):
filelist.extend(findfiles.main(i, verbose))
elif os.path.isfile(i):
filelist.append(i)
else:
txt = 'Input file or folder %s could not be found.' % i
print txt
raise Usage(txt)
dsList = DataSetList(filelist, verbose)
if not dsList:
raise Usage("Nothing to do: post-processing stopped.")
if isNoisy and not isNoiseFree:
dsList = dsList.dictByNoise().get('nzall', DataSetList())
if isNoiseFree and not isNoisy:
dsList = dsList.dictByNoise().get('noiselessall', DataSetList())
# compute maxfuneval values
dict_max_fun_evals = {}
for ds in dsList:
dict_max_fun_evals[ds.dim] = np.max(
(dict_max_fun_evals.setdefault(ds.dim,
0), float(np.max(ds.maxevals))))
if isRLbased is not None:
genericsettings.runlength_based_targets = isRLbased
from bbob_pproc import config
config.target_values(isExpensive, dict_max_fun_evals)
config.config()
if (verbose):
for i in dsList:
if (dict((j, i.instancenumbers.count(j))
for j in set(i.instancenumbers)) !=
inset.instancesOfInterest):
warnings.warn('The data of %s do not list ' % (i) +
'the correct instances ' +
'of function F%d.' % (i.funcId))
dictAlg = dsList.dictByAlg()
if len(dictAlg) > 1:
warnings.warn('Data with multiple algId %s ' % (dictAlg) +
'will be processed together.')
# TODO: in this case, all is well as long as for a given problem
# (given dimension and function) there is a single instance of
# DataSet associated. If there are more than one, the first one only
# will be considered... which is probably not what one would expect.
# TODO: put some errors where this case would be a problem.
# raise Usage?
if isfigure or istab or isrldistr or islogloss:
if not os.path.exists(outputdir):
os.makedirs(outputdir)
if verbose:
print 'Folder %s was created.' % (outputdir)
if isPickled:
dsList.pickle(verbose=verbose)
if isConv:
ppconverrorbars.main(dictAlg, outputdir, verbose)
if isfigure:
print "Scaling figures...",
sys.stdout.flush()
# ERT/dim vs dim.
plt.rc("axes", **inset.rcaxeslarger)
plt.rc("xtick", **inset.rcticklarger)
plt.rc("ytick", **inset.rcticklarger)
plt.rc("font", **inset.rcfontlarger)
plt.rc("legend", **inset.rclegendlarger)
ppfigdim.main(dsList, ppfigdim.values_of_interest, outputdir,
verbose)
plt.rcdefaults()
print_done()
plt.rc("axes", **inset.rcaxes)
plt.rc("xtick", **inset.rctick)
plt.rc("ytick", **inset.rctick)
plt.rc("font", **inset.rcfont)
plt.rc("legend", **inset.rclegend)
if istab:
print "TeX tables...",
sys.stdout.flush()
dictNoise = dsList.dictByNoise()
for noise, sliceNoise in dictNoise.iteritems():
pptable.main(sliceNoise, inset.tabDimsOfInterest, outputdir,
noise, verbose)
print_done()
if isrldistr:
print "ECDF graphs...",
sys.stdout.flush()
dictNoise = dsList.dictByNoise()
if len(dictNoise) > 1:
warnings.warn('Data for functions from both the noisy and '
'non-noisy testbeds have been found. Their '
'results will be mixed in the "all functions" '
'ECDF figures.')
dictDim = dsList.dictByDim()
for dim in inset.rldDimsOfInterest:
try:
sliceDim = dictDim[dim]
except KeyError:
continue
pprldistr.main(sliceDim, True, outputdir, 'all', verbose)
dictNoise = sliceDim.dictByNoise()
for noise, sliceNoise in dictNoise.iteritems():
pprldistr.main(sliceNoise, True, outputdir, '%s' % noise,
verbose)
dictFG = sliceDim.dictByFuncGroup()
for fGroup, sliceFuncGroup in dictFG.items():
pprldistr.main(sliceFuncGroup, True, outputdir,
'%s' % fGroup, verbose)
pprldistr.fmax = None # Resetting the max final value
pprldistr.evalfmax = None # Resetting the max #fevalsfactor
print_done()
if islogloss:
print "ERT loss ratio figures and tables...",
sys.stdout.flush()
for ng, sliceNoise in dsList.dictByNoise().iteritems():
if ng == 'noiselessall':
testbed = 'noiseless'
elif ng == 'nzall':
testbed = 'noisy'
txt = ("Please input crafting effort value " +
"for %s testbed:\n CrE = " % testbed)
CrE = inputCrE
while CrE is None:
try:
CrE = float(raw_input(txt))
except (SyntaxError, NameError, ValueError):
print "Float value required."
dictDim = sliceNoise.dictByDim()
for d in inset.rldDimsOfInterest:
try:
sliceDim = dictDim[d]
except KeyError:
continue
info = '%s' % ng
pplogloss.main(sliceDim,
CrE,
True,
outputdir,
info,
verbose=verbose)
pplogloss.generateTable(sliceDim,
CrE,
outputdir,
info,
verbose=verbose)
for fGroup, sliceFuncGroup in sliceDim.dictByFuncGroup(
).iteritems():
info = '%s' % fGroup
pplogloss.main(sliceFuncGroup,
CrE,
True,
outputdir,
info,
verbose=verbose)
pplogloss.evalfmax = None # Resetting the max #fevalsfactor
print_done()
latex_commands_file = os.path.join(
outputdir.split(os.sep)[0], 'bbob_pproc_commands.tex')
prepend_to_file(latex_commands_file, [
'\\providecommand{\\bbobloglosstablecaption}[1]{',
pplogloss.table_caption, '}'
])
prepend_to_file(latex_commands_file, [
'\\providecommand{\\bbobloglossfigurecaption}[1]{',
pplogloss.figure_caption, '}'
])
prepend_to_file(
latex_commands_file,
[
'\\providecommand{\\bbobpprldistrlegend}[1]{',
pprldistr.caption_single(
np.max([
val / dim
for dim, val in dict_max_fun_evals.iteritems()
])
), # depends on the config setting, should depend on maxfevals
'}'
])
prepend_to_file(latex_commands_file, [
'\\providecommand{\\bbobppfigdimlegend}[1]{',
ppfigdim.scaling_figure_caption(), '}'
])
prepend_to_file(latex_commands_file, [
'\\providecommand{\\bbobpptablecaption}[1]{',
pptable.table_caption, '}'
])
prepend_to_file(latex_commands_file,
['\\providecommand{\\algfolder}{}'
]) # is overwritten in rungeneric.py
prepend_to_file(latex_commands_file, [
'\\providecommand{\\algname}{' +
(str_to_latex(strip_pathname(args[0]))
if len(args) == 1 else str_to_latex(dsList[0].algId)) + '{}}'
])
if isfigure or istab or isrldistr or islogloss:
print "Output data written to folder %s" % outputdir
plt.rcdefaults()
def main(argv=None):
"""Generate python-formatted data from raw BBOB experimental data.
The raw experimental data (files with the extension :file:`info`
pointing to files with extension :file:`dat` and :file:`tdat`) are
post-processed and stored in a more condensed way as files with the
extension :file:`pickle`.
Supposing the raw data are stored in folder :file:`mydata`, the new
pickle files will be put in folder :file:`mydata-pickle`.
:keyword list argv: strings containing options and arguments. If not
provided, sys.argv is accessed.
*argv* should list either names of info files or folders containing
info files.
Furthermore, *argv* can begin with, in any order, facultative option
flags listed below.
-h, --help
display this message
:exception Usage: Gives back a usage message.
Examples:
* Calling the dataoutput.py interface from the command line::
$ python bbob_pproc/dataoutput.py experiment2/*.info
* Loading this package and calling the main from the command line
(requires that the path to this package is in the search path)::
$ python -m bbob_pproc.dataoutput -h
This will print out this help message.
* From the python interactive shell (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.dataoutput.main('folder1')
"""
if argv is None:
argv = sys.argv[1:]
try:
try:
opts, args = getopt.getopt(argv, "h", ["help"])
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
verbose = False
#Process options
for o, a in opts:
if o in ("-h", "--help"):
usage()
sys.exit()
else:
assert False, "unhandled option"
if (not verbose):
warnings.simplefilter('ignore')
dsList = DataSetList(args)
outputPickle(dsList, verbose=True)
def main(argv=None):
r"""Routine for post-processing COCO data from two algorithms.
Provided with some data, this routine outputs figure and TeX files
in a folder needed for the compilation of the provided LaTeX templates
for comparing two algorithms (``*cmp.tex`` or ``*2*.tex``).
The used template file needs to be edited so that the command
``\bbobdatapath`` points to the output folder created by this routine.
The output files will contain performance tables, performance
scaling figures and empirical cumulative distribution figures. On
subsequent executions, new files will be added to the output folder,
overwriting existing older files in the process.
Keyword arguments:
*argv* -- list of strings containing options and arguments. If not
given, sys.argv is accessed.
*argv* must list folders containing BBOB data files. Each of these
folders should correspond to the data of ONE algorithm.
Furthermore, argv can begin with, in any order, facultative option
flags listed below.
-h, --help
displays this message.
-v, --verbose
verbose mode, prints out operations.
-o OUTPUTDIR, --output-dir=OUTPUTDIR
changes the default output directory (:file:`ppdata`) to
:file:`OUTPUTDIR`
--noise-free, --noisy
processes only part of the data.
--settings=SETTING
changes the style of the output figures and tables. At the
moment only the only differences are in the colors of the
output figures. SETTING can be either "grayscale", "color"
or "black-white". The default setting is "color".
--fig-only, --rld-only, --tab-only, --sca-only
these options can be used to output respectively the ERT
graphs figures, run length distribution figures or the
comparison tables scatter plot figures only. Any combination
of these options results in no output.
--conv
if this option is chosen, additionally convergence
plots for each function and algorithm are generated.
--rld-single-fcts
generate also runlength distribution figures for each
single function.
--expensive
runlength-based f-target values and fixed display limits,
useful with comparatively small budgets. By default the
setting is based on the budget used in the data.
--not-expensive
expensive setting off.
--svg
generate also the svg figures which are used in html files
Exceptions raised:
*Usage* -- Gives back a usage message.
Examples:
* Calling the rungeneric2.py interface from the command line::
$ python bbob_pproc/rungeneric2.py -v Alg0-baseline Alg1-of-interest
will post-process the data from folders :file:`Alg0-baseline` and
:file:`Alg1-of-interest`, the former containing data for the
reference algorithm (zero-th) and the latter data for the
algorithm of concern (first). The results will be output in the
default output folder. The ``-v`` option adds verbosity.
* From the python interpreter (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.rungeneric2.main('-o outputfolder PSO DEPSO'.split())
This will execute the post-processing on the data found in folder
:file:`PSO` and :file:`DEPSO`. The ``-o`` option changes the output
folder from the default to :file:`outputfolder`.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
global ftarget
try:
try:
opts, args = getopt.getopt(argv, genericsettings.shortoptlist, genericsettings.longoptlist)
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
#Process options
outputdir = genericsettings.outputdir
for o, a in opts:
if o in ("-v", "--verbose"):
genericsettings.verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-o", "--output-dir"):
outputdir = a
elif o == "--fig-only":
genericsettings.isRLDistr = False
genericsettings.isTab = False
genericsettings.isScatter = False
elif o == "--rld-only":
genericsettings.isFig = False
genericsettings.isTab = False
genericsettings.isScatter = False
elif o == "--tab-only":
genericsettings.isFig = False
genericsettings.isRLDistr = False
genericsettings.isScatter = False
elif o == "--sca-only":
genericsettings.isFig = False
genericsettings.isRLDistr = False
genericsettings.isTab = False
elif o == "--noisy":
genericsettings.isNoisy = True
elif o == "--noise-free":
genericsettings.isNoiseFree = True
elif o == "--settings":
genericsettings.inputsettings = a
elif o == "--conv":
genericsettings.isConv = True
elif o == "--rld-single-fcts":
genericsettings.isRldOnSingleFcts = True
elif o == "--runlength-based":
genericsettings.runlength_based_targets = True
elif o == "--expensive":
genericsettings.isExpensive = True # comprises runlength-based
elif o == "--not-expensive":
genericsettings.isExpensive = False
elif o == "--svg":
genericsettings.generate_svg_files = True
elif o == "--los-only":
warnings.warn("option --los-only will have no effect with rungeneric2.py")
elif o == "--crafting-effort=":
warnings.warn("option --crafting-effort will have no effect with rungeneric2.py")
elif o in ("-p", "--pickle"):
warnings.warn("option --pickle will have no effect with rungeneric2.py")
else:
assert False, "unhandled option"
# from bbob_pproc import bbob2010 as inset # input settings
if genericsettings.inputsettings == "color":
from bbob_pproc import genericsettings as inset # input settings
config.config()
elif genericsettings.inputsettings == "grayscale": # probably very much obsolete
from bbob_pproc import grayscalesettings as inset # input settings
elif genericsettings.inputsettings == "black-white": # probably very much obsolete
from bbob_pproc import bwsettings as inset # input settings
else:
txt = ('Settings: %s is not an appropriate ' % genericsettings.inputsettings
+ 'argument for input flag "--settings".')
raise Usage(txt)
if (not genericsettings.verbose):
warnings.simplefilter('module')
warnings.simplefilter('ignore')
print ("Post-processing will generate comparison " +
"data in folder %s" % outputdir)
print " this might take several minutes."
dsList, sortedAlgs, dictAlg = processInputArgs(args, verbose=genericsettings.verbose)
if 1 < 3 and len(sortedAlgs) != 2:
raise ValueError('rungeneric2.py needs exactly two algorithms to compare, found: '
+ str(sortedAlgs)
+ '\n use rungeneric.py (or rungenericmany.py) to compare more algorithms. ')
if not dsList:
sys.exit()
for i in dictAlg:
if genericsettings.isNoisy and not genericsettings.isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get('nzall', DataSetList())
if genericsettings.isNoiseFree and not genericsettings.isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get('noiselessall', DataSetList())
for i in dsList:
if i.dim not in genericsettings.dimensions_to_display:
continue
if (dict((j, i.instancenumbers.count(j)) for j in set(i.instancenumbers)) <
inset.instancesOfInterest):
warnings.warn('The data of %s do not list ' %(i) +
'the correct instances ' +
'of function F%d.' %(i.funcId))
if len(sortedAlgs) < 2:
raise Usage('Expect data from two different algorithms, could ' +
'only find one.')
elif len(sortedAlgs) > 2:
warnings.warn('Data from folders: %s ' % (sortedAlgs) +
'were found, the first two will be processed.')
# Group by algorithm
dsList0 = dictAlg[sortedAlgs[0]]
if not dsList0:
raise Usage('Could not find data for algorithm %s.' % (sortedAlgs[0]))
dsList1 = dictAlg[sortedAlgs[1]]
if not dsList1:
raise Usage('Could not find data for algorithm %s.' % (sortedAlgs[0]))
# get the name of each algorithm from the input arguments
tmppath0, alg0name = os.path.split(sortedAlgs[0].rstrip(os.sep))
tmppath1, alg1name = os.path.split(sortedAlgs[1].rstrip(os.sep))
for i in dsList0:
i.algId = alg0name
for i in dsList1:
i.algId = alg1name
# compute maxfuneval values
dict_max_fun_evals1 = {}
dict_max_fun_evals2 = {}
for ds in dsList0:
dict_max_fun_evals1[ds.dim] = np.max((dict_max_fun_evals1.setdefault(ds.dim, 0), float(np.max(ds.maxevals))))
for ds in dsList1:
dict_max_fun_evals2[ds.dim] = np.max((dict_max_fun_evals2.setdefault(ds.dim, 0), float(np.max(ds.maxevals))))
config.target_values(genericsettings.isExpensive, {1: min([max([val/dim for dim, val in dict_max_fun_evals1.iteritems()]),
max([val/dim for dim, val in dict_max_fun_evals2.iteritems()])]
)})
config.config()
######################### Post-processing #############################
if genericsettings.isFig or genericsettings.isRLDistr or genericsettings.isTab or genericsettings.isScatter:
if not os.path.exists(outputdir):
os.mkdir(outputdir)
if genericsettings.verbose:
print 'Folder %s was created.' % (outputdir)
# prepend the algorithm name command to the tex-command file
abc = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
lines = []
for i, alg in enumerate(args):
lines.append('\\providecommand{\\algorithm' + abc[i] + '}{' +
str_to_latex(strip_pathname1(alg)) + '}')
prepend_to_file(os.path.join(outputdir,
'bbob_pproc_commands.tex'), lines, 1000,
'bbob_proc_commands.tex truncated, consider removing the file before the text run'
)
# Check whether both input arguments list noisy and noise-free data
dictFN0 = dsList0.dictByNoise()
dictFN1 = dsList1.dictByNoise()
k0 = set(dictFN0.keys())
k1 = set(dictFN1.keys())
symdiff = k1 ^ k0 # symmetric difference
if symdiff:
tmpdict = {}
for i, noisegrp in enumerate(symdiff):
if noisegrp == 'nzall':
tmp = 'noisy'
elif noisegrp == 'noiselessall':
tmp = 'noiseless'
if dictFN0.has_key(noisegrp):
tmp2 = sortedAlgs[0]
elif dictFN1.has_key(noisegrp):
tmp2 = sortedAlgs[1]
tmpdict.setdefault(tmp2, []).append(tmp)
txt = []
for i, j in tmpdict.iteritems():
txt.append('Only input folder %s lists %s data.'
% (i, ' and '.join(j)))
raise Usage('Data Mismatch: \n ' + ' '.join(txt)
+ '\nTry using --noise-free or --noisy flags.')
if genericsettings.isFig:
plt.rc("axes", **inset.rcaxeslarger)
plt.rc("xtick", **inset.rcticklarger)
plt.rc("ytick", **inset.rcticklarger)
plt.rc("font", **inset.rcfontlarger)
plt.rc("legend", **inset.rclegendlarger)
plt.rc('pdf', fonttype = 42)
ppfig2.main(dsList0, dsList1, ppfig2_ftarget,
outputdir, genericsettings.verbose)
print "log ERT1/ERT0 vs target function values done."
plt.rc("axes", **inset.rcaxes)
plt.rc("xtick", **inset.rctick)
plt.rc("ytick", **inset.rctick)
plt.rc("font", **inset.rcfont)
plt.rc("legend", **inset.rclegend)
plt.rc('pdf', fonttype = 42)
if genericsettings.isRLDistr:
if len(dictFN0) > 1 or len(dictFN1) > 1:
warnings.warn('Data for functions from both the noisy and ' +
'non-noisy testbeds have been found. Their ' +
'results will be mixed in the "all functions" ' +
'ECDF figures.')
dictDim0 = dsList0.dictByDim()
dictDim1 = dsList1.dictByDim()
# ECDFs of ERT ratios
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
if dim in inset.rldDimsOfInterest:
# ECDF for all functions altogether
try:
pprldistr2.main(dictDim0[dim], dictDim1[dim], dim,
inset.rldValsOfInterest,
outputdir,
'%02dD_all' % dim,
genericsettings.verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.'
% (dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr2.main(dictFG1[fGroup], dictFG0[fGroup], dim,
inset.rldValsOfInterest,
outputdir,
'%02dD_%s' % (dim, fGroup),
genericsettings.verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr2.main(dictFN1[fGroup], dictFN0[fGroup], dim,
inset.rldValsOfInterest,
outputdir,
'%02dD_%s' % (dim, fGroup),
genericsettings.verbose)
prepend_to_file(os.path.join(outputdir,
'bbob_pproc_commands.tex'),
['\\providecommand{\\bbobpprldistrlegendtwo}[1]{',
pprldistr.caption_two(), # depends on the config setting, should depend on maxfevals
'}'
])
print "ECDF runlength ratio graphs done."
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
pprldistr.fmax = None #Resetting the max final value
pprldistr.evalfmax = None #Resetting the max #fevalsfactor
# ECDFs of all functions altogether
if dim in inset.rldDimsOfInterest:
try:
pprldistr.comp(dictDim1[dim], dictDim0[dim],
inset.rldValsOfInterest, # TODO: let rldVals... possibly be RL-based targets
True,
outputdir, 'all', genericsettings.verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.'
% (dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr.comp(dictFG1[fGroup], dictFG0[fGroup],
inset.rldValsOfInterest, True,
outputdir,
'%s' % fGroup, genericsettings.verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr.comp(dictFN1[fGroup], dictFN0[fGroup],
inset.rldValsOfInterest, True,
outputdir,
'%s' % fGroup, genericsettings.verbose)
if genericsettings.isRldOnSingleFcts: # copy-paste from above, here for each function instead of function groups
# ECDFs for each function
pprldmany.all_single_functions(dictAlg, sortedAlgs,
outputdir, genericsettings.verbose)
print "ECDF runlength graphs done."
if genericsettings.isConv:
ppconverrorbars.main(dictAlg, outputdir, genericsettings.verbose)
if genericsettings.isScatter:
if genericsettings.runlength_based_targets:
ppscatter.targets = ppscatter.runlength_based_targets
ppscatter.main(dsList1, dsList0, outputdir,
verbose=genericsettings.verbose)
prepend_to_file(os.path.join(outputdir,
'bbob_pproc_commands.tex'),
['\\providecommand{\\bbobppscatterlegend}[1]{',
ppscatter.figure_caption(),
'}'
])
replace_in_file(os.path.join(outputdir, genericsettings.two_algorithm_file_name + '.html'), '##bbobppscatterlegend##', ppscatter.figure_caption_html())
print "Scatter plots done."
if genericsettings.isTab:
dictNG0 = dsList0.dictByNoise()
dictNG1 = dsList1.dictByNoise()
for nGroup in set(dictNG0.keys()) & set(dictNG1.keys()):
# split table in as many as necessary
dictFunc0 = dictNG0[nGroup].dictByFunc()
dictFunc1 = dictNG1[nGroup].dictByFunc()
funcs = list(set(dictFunc0.keys()) & set(dictFunc1.keys()))
if len(funcs) > 24:
funcs.sort()
nbgroups = int(numpy.ceil(len(funcs)/24.))
def split_seq(seq, nbgroups):
newseq = []
splitsize = 1.0/nbgroups*len(seq)
for i in range(nbgroups):
newseq.append(seq[int(round(i*splitsize)):int(round((i+1)*splitsize))])
return newseq
groups = split_seq(funcs, nbgroups)
# merge
group0 = []
group1 = []
for i, g in enumerate(groups):
tmp0 = DataSetList()
tmp1 = DataSetList()
for f in g:
tmp0.extend(dictFunc0[f])
tmp1.extend(dictFunc1[f])
group0.append(tmp0)
group1.append(tmp1)
for i, g in enumerate(zip(group0, group1)):
pptable2.main(g[0], g[1], inset.tabDimsOfInterest,
outputdir,
'%s%d' % (nGroup, i), genericsettings.verbose)
else:
if 11 < 3: # future handling:
dictFunc0 = dsList0.dictByFunc()
dictFunc1 = dsList1.dictByFunc()
funcs = list(set(dictFunc0.keys()) & set(dictFunc1.keys()))
funcs.sort()
# nbgroups = int(numpy.ceil(len(funcs)/testbedsettings.numberOfFunctions))
# pptable2.main(dsList0, dsList1,
# testbedsettings.tabDimsOfInterest, outputdir,
# '%s' % (testbedsettings.testbedshortname), genericsettings.verbose)
else:
pptable2.main(dictNG0[nGroup], dictNG1[nGroup],
inset.tabDimsOfInterest,
outputdir,
'%s' % (nGroup), genericsettings.verbose)
if isinstance(pptable2.targetsOfInterest, pproc.RunlengthBasedTargetValues):
prepend_to_file(os.path.join(outputdir,
'bbob_pproc_commands.tex'),
['\\providecommand{\\bbobpptablestwolegend}[1]{',
pptable2.table_caption_expensive,
'}'
])
else:
prepend_to_file(os.path.join(outputdir,
'bbob_pproc_commands.tex'),
['\\providecommand{\\bbobpptablestwolegend}[1]{',
pptable2.table_caption,
'}'
])
htmlFileName = os.path.join(outputdir, genericsettings.two_algorithm_file_name + '.html')
key = '##bbobpptablestwolegendexpensive##' if isinstance(pptable2.targetsOfInterest, pproc.RunlengthBasedTargetValues) else '##bbobpptablestwolegend##'
replace_in_file(htmlFileName, '##bbobpptablestwolegend##', htmldesc.getValue(key))
alg0 = set(i[0] for i in dsList0.dictByAlg().keys()).pop().replace(genericsettings.extraction_folder_prefix, '')[0:3]
alg1 = set(i[0] for i in dsList1.dictByAlg().keys()).pop().replace(genericsettings.extraction_folder_prefix, '')[0:3]
replace_in_file(htmlFileName, 'algorithmAshort', alg0)
replace_in_file(htmlFileName, 'algorithmBshort', alg1)
for i, alg in enumerate(args):
replace_in_file(htmlFileName, 'algorithm' + abc[i], str_to_latex(strip_pathname1(alg)))
print "Tables done."
if genericsettings.isScaleUp:
plt.rc("axes", labelsize=20, titlesize=24)
plt.rc("xtick", labelsize=20)
plt.rc("ytick", labelsize=20)
plt.rc("font", size=20)
plt.rc("legend", fontsize=20)
plt.rc('pdf', fonttype = 42)
if genericsettings.runlength_based_targets:
ftarget = RunlengthBasedTargetValues([target_runlength]) # TODO: make this more variable but also consistent
ppfigs.main(dictAlg, genericsettings.two_algorithm_file_name, sortedAlgs, ftarget,
outputdir, genericsettings.verbose)
plt.rcdefaults()
print "Scaling figures done."
if genericsettings.isFig or genericsettings.isRLDistr or genericsettings.isTab or genericsettings.isScatter or genericsettings.isScaleUp:
print "Output data written to folder %s" % outputdir
plt.rcdefaults()
for i, alg in enumerate(args):
lines.append('\\providecommand{\\algorithm' + abc[i] + '}{' +
str_to_latex(strip_pathname2(alg)) + '}')
prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'),
lines, 5000,
'bbob_proc_commands.tex truncated, consider removing the file before the text run'
)
dsList, sortedAlgs, dictAlg = processInputArgs(args, verbose=verbose)
if not dsList:
sys.exit()
for i in dictAlg:
if isNoisy and not isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get('nzall', DataSetList())
if isNoiseFree and not isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get('noiselessall', DataSetList())
# compute maxfuneval values
# TODO: we should rather take min_algorithm max_evals
dict_max_fun_evals = {}
for ds in dsList:
dict_max_fun_evals[ds.dim] = numpy.max((dict_max_fun_evals.setdefault(ds.dim, 0), float(numpy.max(ds.maxevals))))
if isRLbased is not None:
genericsettings.runlength_based_targets = isRLbased
# set target values
from bbob_pproc import config
def main(argv=None):
r"""Routine for post-processing COCO data from two algorithms.
Provided with some data, this routine outputs figure and TeX files
in a folder needed for the compilation of latex document
:file:`template2XXX.tex` or :file:`noisytemplate2XXX.tex`, where
:file:`XXX` is either :file:`ecj` or :file:`generic`. The template
file needs to be edited so that the command ``\bbobdatapath`` points
to the output folder.
These output files will contain performance tables, performance
scaling figures and empirical cumulative distribution figures. On
subsequent executions, new files will be added to the output folder,
overwriting existing older files in the process.
Keyword arguments:
*argv* -- list of strings containing options and arguments. If not
given, sys.argv is accessed.
*argv* must list folders containing BBOB data files. Each of these
folders should correspond to the data of ONE algorithm.
Furthermore, argv can begin with, in any order, facultative option
flags listed below.
-h, --help
displays this message.
-v, --verbose
verbose mode, prints out operations.
-o OUTPUTDIR, --output-dir=OUTPUTDIR
changes the default output directory (:file:`ppdata`) to
:file:`OUTPUTDIR`
--noise-free, --noisy
processes only part of the data.
--settings=SETTING
changes the style of the output figures and tables. At the
moment only the only differences are in the colors of the
output figures. SETTING can be either "grayscale", "color"
or "black-white". The default setting is "color".
--fig-only, --rld-only, --tab-only, --sca-only
these options can be used to output respectively the ERT
graphs figures, run length distribution figures or the
comparison tables scatter plot figures only. Any combination
of these options results in no output.
--conv
if this option is chosen, additionally convergence
plots for each function and algorithm are generated.
Exceptions raised:
*Usage* -- Gives back a usage message.
Examples:
* Calling the rungeneric2.py interface from the command line::
$ python bbob_pproc/rungeneric2.py -v Alg0-baseline Alg1-of-interest
will post-process the data from folders :file:`Alg0-baseline` and
:file:`Alg1-of-interest`, the former containing data for the
reference algorithm (zero-th) and the latter data for the
algorithm of concern (first). The results will be output in the
default output folder. The ``-v`` option adds verbosity.
* From the python interpreter (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.rungeneric2.main('-o outputfolder PSO DEPSO'.split())
This will execute the post-processing on the data found in folder
:file:`PSO` and :file:`DEPSO`. The ``-o`` option changes the output
folder from the default to :file:`outputfolder`.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
try:
try:
opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
isfigure = True
isrldistr = True
istable = True
isscatter = True
isscaleup = True
isNoisy = False
isNoiseFree = False
verbose = False
outputdir = 'ppdata'
inputsettings = 'color'
isConv = False
#Process options
for o, a in opts:
if o in ("-v", "--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-o", "--output-dir"):
outputdir = a
#elif o in ("-s", "--style"):
# inputsettings = a
elif o == "--fig-only":
isrldistr = False
istable = False
isscatter = False
elif o == "--rld-only":
isfigure = False
istable = False
isscatter = False
elif o == "--tab-only":
isfigure = False
isrldistr = False
isscatter = False
elif o == "--sca-only":
isfigure = False
isrldistr = False
istable = False
elif o == "--noisy":
isNoisy = True
elif o == "--noise-free":
isNoiseFree = True
elif o == "--settings":
inputsettings = a
elif o == "--conv":
isConv = True
else:
assert False, "unhandled option"
# from bbob_pproc import bbob2010 as inset # input settings
if inputsettings == "color":
from bbob_pproc import config, genericsettings as inset # input settings
config.config()
elif inputsettings == "grayscale":
from bbob_pproc import grayscalesettings as inset # input settings
elif inputsettings == "black-white":
from bbob_pproc import bwsettings as inset # input settings
else:
txt = ('Settings: %s is not an appropriate ' % inputsettings +
'argument for input flag "--settings".')
raise Usage(txt)
if (not verbose):
warnings.simplefilter('module')
warnings.simplefilter('ignore')
print("Post-processing will generate comparison " +
"data in folder %s" % outputdir)
print " this might take several minutes."
dsList, sortedAlgs, dictAlg = processInputArgs(args, verbose=verbose)
if 1 < 3 and len(sortedAlgs) != 2:
raise ValueError(
'rungeneric2.py needs exactly two algorithms to compare, found: '
+ str(sortedAlgs) +
'\n use rungeneric.py (or rungenericmany.py) to compare more algorithms. '
)
if not dsList:
sys.exit()
for i in dictAlg:
if isNoisy and not isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'nzall', DataSetList())
if isNoiseFree and not isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'noiselessall', DataSetList())
for i in dsList:
if i.dim not in genericsettings.dimensions_to_display:
continue
if (dict((j, i.instancenumbers.count(j))
for j in set(i.instancenumbers)) <
inset.instancesOfInterest):
warnings.warn('The data of %s do not list ' % (i) +
'the correct instances ' + 'of function F%d.' %
(i.funcId))
if len(sortedAlgs) < 2:
raise Usage('Expect data from two different algorithms, could ' +
'only find one.')
elif len(sortedAlgs) > 2:
warnings.warn('Data from folders: %s ' % (sortedAlgs) +
'were found, the first two will be processed.')
# Group by algorithm
dsList0 = dictAlg[sortedAlgs[0]]
if not dsList0:
raise Usage('Could not find data for algorithm %s.' %
(sortedAlgs[0]))
dsList1 = dictAlg[sortedAlgs[1]]
if not dsList1:
raise Usage('Could not find data for algorithm %s.' %
(sortedAlgs[0]))
# get the name of each algorithm from the input arguments
tmppath0, alg0name = os.path.split(sortedAlgs[0].rstrip(os.sep))
tmppath1, alg1name = os.path.split(sortedAlgs[1].rstrip(os.sep))
for i in dsList0:
i.algId = alg0name
for i in dsList1:
i.algId = alg1name
######################### Post-processing #############################
if isfigure or isrldistr or istable or isscatter:
if not os.path.exists(outputdir):
os.mkdir(outputdir)
if verbose:
print 'Folder %s was created.' % (outputdir)
# prepend the algorithm name command to the tex-command file
abc = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
lines = []
for i, alg in enumerate(args):
lines.append('\\providecommand{\\algorithm' + abc[i] + '}{' +
str_to_latex(strip_pathname(alg)) + '}')
prepend_to_file(
os.path.join(outputdir,
'bbob_pproc_commands.tex'), lines, 1000,
'bbob_proc_commands.tex truncated, consider removing the file before the text run'
)
# Check whether both input arguments list noisy and noise-free data
dictFN0 = dsList0.dictByNoise()
dictFN1 = dsList1.dictByNoise()
k0 = set(dictFN0.keys())
k1 = set(dictFN1.keys())
symdiff = k1 ^ k0 # symmetric difference
if symdiff:
tmpdict = {}
for i, noisegrp in enumerate(symdiff):
if noisegrp == 'nzall':
tmp = 'noisy'
elif noisegrp == 'noiselessall':
tmp = 'noiseless'
if dictFN0.has_key(noisegrp):
tmp2 = sortedAlgs[0]
elif dictFN1.has_key(noisegrp):
tmp2 = sortedAlgs[1]
tmpdict.setdefault(tmp2, []).append(tmp)
txt = []
for i, j in tmpdict.iteritems():
txt.append('Only input folder %s lists %s data.' %
(i, ' and '.join(j)))
raise Usage('Data Mismatch: \n ' + ' '.join(txt) +
'\nTry using --noise-free or --noisy flags.')
if isfigure:
plt.rc("axes", **inset.rcaxeslarger)
plt.rc("xtick", **inset.rcticklarger)
plt.rc("ytick", **inset.rcticklarger)
plt.rc("font", **inset.rcfontlarger)
plt.rc("legend", **inset.rclegendlarger)
ppfig2.main(dsList0, dsList1, ftarget, outputdir, verbose)
print "log ERT1/ERT0 vs target function values done."
plt.rc("axes", **inset.rcaxes)
plt.rc("xtick", **inset.rctick)
plt.rc("ytick", **inset.rctick)
plt.rc("font", **inset.rcfont)
plt.rc("legend", **inset.rclegend)
if isrldistr:
if len(dictFN0) > 1 or len(dictFN1) > 1:
warnings.warn('Data for functions from both the noisy and ' +
'non-noisy testbeds have been found. Their ' +
'results will be mixed in the "all functions" ' +
'ECDF figures.')
dictDim0 = dsList0.dictByDim()
dictDim1 = dsList1.dictByDim()
# ECDFs of ERT ratios
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
if dim in inset.rldDimsOfInterest:
# ECDF for all functions altogether
try:
pprldistr2.main(dictDim0[dim], dictDim1[dim],
inset.rldValsOfInterest, outputdir,
'%02dD_all' % dim, verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.' %
(dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr2.main(dictFG1[fGroup], dictFG0[fGroup],
inset.rldValsOfInterest, outputdir,
'%02dD_%s' % (dim, fGroup), verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr2.main(dictFN1[fGroup], dictFN0[fGroup],
inset.rldValsOfInterest, outputdir,
'%02dD_%s' % (dim, fGroup), verbose)
print "ECDF runlength ratio graphs done."
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
pprldistr.fmax = None #Resetting the max final value
pprldistr.evalfmax = None #Resetting the max #fevalsfactor
# ECDFs of all functions altogether
if dim in inset.rldDimsOfInterest:
try:
pprldistr.comp(
dictDim1[dim], dictDim0[dim],
inset.rldValsOfInterest if isinstance(
inset.rldValsOfInterest, TargetValues) else
TargetValues(inset.rldValsOfInterest), True,
outputdir, 'all', verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.' %
(dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr.comp(
dictFG1[fGroup], dictFG0[fGroup],
inset.rldValsOfInterest if isinstance(
inset.rldValsOfInterest, TargetValues) else
TargetValues(inset.rldValsOfInterest), True,
outputdir, '%s' % fGroup, verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr.comp(
dictFN1[fGroup], dictFN0[fGroup],
inset.rldValsOfInterest if isinstance(
inset.rldValsOfInterest, TargetValues) else
TargetValues(inset.rldValsOfInterest), True,
outputdir, '%s' % fGroup, verbose)
print "ECDF runlength graphs done."
if isConv:
ppconverrorbars.main(dictAlg, outputdir, verbose)
if istable:
dictNG0 = dsList0.dictByNoise()
dictNG1 = dsList1.dictByNoise()
for nGroup in set(dictNG0.keys()) & set(dictNG1.keys()):
# split table in as many as necessary
dictFunc0 = dictNG0[nGroup].dictByFunc()
dictFunc1 = dictNG1[nGroup].dictByFunc()
funcs = list(set(dictFunc0.keys()) & set(dictFunc1.keys()))
if len(funcs) > 24:
funcs.sort()
nbgroups = int(numpy.ceil(len(funcs) / 24.))
def split_seq(seq, nbgroups):
newseq = []
splitsize = 1.0 / nbgroups * len(seq)
for i in range(nbgroups):
newseq.append(
seq[int(round(i * splitsize)
):int(round((i + 1) * splitsize))])
return newseq
groups = split_seq(funcs, nbgroups)
# merge
group0 = []
group1 = []
for i, g in enumerate(groups):
tmp0 = DataSetList()
tmp1 = DataSetList()
for f in g:
tmp0.extend(dictFunc0[f])
tmp1.extend(dictFunc1[f])
group0.append(tmp0)
group1.append(tmp1)
for i, g in enumerate(zip(group0, group1)):
pptable2.main(g[0], g[1], inset.tabDimsOfInterest,
outputdir, '%s%d' % (nGroup, i), verbose)
else:
if 11 < 3: # future handling:
dictFunc0 = dsList0.dictByFunc()
dictFunc1 = dsList1.dictByFunc()
funcs = list(
set(dictFunc0.keys()) & set(dictFunc1.keys()))
funcs.sort()
# nbgroups = int(numpy.ceil(len(funcs)/testbedsettings.numberOfFunctions))
# pptable2.main(dsList0, dsList1,
# testbedsettings.tabDimsOfInterest, outputdir,
# '%s' % (testbedsettings.testbedshortname), verbose)
else:
pptable2.main(dictNG0[nGroup], dictNG1[nGroup],
inset.tabDimsOfInterest, outputdir,
'%s' % (nGroup), verbose)
prepend_to_file(os.path.join(
outputdir, 'bbob_pproc_commands.tex'), [
'\\providecommand{\\bbobpptablestwolegend}[1]{',
pptable2.figure_legend, '}'
])
print "Tables done."
if isscatter:
ppscatter.main(dsList0, dsList1, outputdir, verbose=verbose)
prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'),
[
'\\providecommand{\\bbobppscatterlegend}[1]{',
ppscatter.figure_legend, '}'
])
print "Scatter plots done."
if isscaleup:
plt.rc("axes", labelsize=20, titlesize=24)
plt.rc("xtick", labelsize=20)
plt.rc("ytick", labelsize=20)
plt.rc("font", size=20)
plt.rc("legend", fontsize=20)
ppfigs.main(dictAlg, sortedAlgs, ftarget, outputdir, verbose)
plt.rcdefaults()
print "Scaling figures done."
if isfigure or isrldistr or istable or isscatter or isscaleup:
print "Output data written to folder %s" % outputdir
plt.rcdefaults()
def main(argv=None):
r"""Post-processing COCO data of a single algorithm.
Provided with some data, this routine outputs figure and TeX files
in a folder needed for the compilation of latex document
:file:`template1XXX.tex` or :file:`noisytemplate1XXX.tex`, where
:file:`XXX` is either :file:`ecj` or :file:`generic`. The template
file needs to be edited so that the commands
``\bbobdatapath`` and ``\algfolder`` point to the output folder.
These output files will contain performance tables, performance
scaling figures and empirical cumulative distribution figures. On
subsequent executions, new files will be added to the output folder,
overwriting existing older files in the process.
Keyword arguments:
*argv* -- list of strings containing options and arguments. If not
given, sys.argv is accessed.
*argv* should list either names of :file:`info` files or folders
containing :file:`info` files. argv can also contain post-processed
:file:`pickle` files generated by this routine. Furthermore, *argv*
can begin with, in any order, facultative option flags listed below.
-h, --help
displays this message.
-v, --verbose
verbose mode, prints out all operations.
-p, --pickle
generates pickle post processed data files.
-o OUTPUTDIR, --output-dir=OUTPUTDIR
changes the default output directory (:file:`ppdata`) to
:file:`OUTPUTDIR`.
--crafting-effort=VALUE
sets the crafting effort to VALUE (float). Otherwise the
default value of 0. will be used.
--noise-free, --noisy
processes only part of the data.
--settings=SETTINGS
changes the style of the output figures and tables. At the
moment the only differences are in the colors of the output
figures. SETTINGS can be either "grayscale", "color" or
"black-white". The default setting is "color".
--tab-only, --fig-only, --rld-only, --los-only
these options can be used to output respectively the TeX
tables, convergence and ERTs graphs figures, run length
distribution figures, ERT loss ratio figures only. A
combination of any two of these options results in no
output.
--conv
if this option is chosen, additionally convergence plots
for each function and algorithm are generated.
--expensive
runlength-based f-target values and fixed display limits,
useful with comparatively small budgets. By default the
setting is based on the budget used in the data.
--not-expensive
expensive setting off.
--runlength-based
runlength-based f-target values, such that the
"level of difficulty" is similar for all functions.
Exceptions raised:
*Usage* -- Gives back a usage message.
Examples:
* Calling the rungeneric1.py interface from the command line::
$ python bbob_pproc/rungeneric1.py -v experiment1
will post-process the folder experiment1 and all its containing
data, base on the .info files found in the folder. The result will
appear in the default output folder. The -v option adds verbosity. ::
$ python bbob_pproc/rungeneric1.py -o exp2 experiment2/*.info
This will execute the post-processing on the info files found in
:file:`experiment2`. The result will be located in the alternative
location :file:`exp2`.
* Loading this package and calling the main from the command line
(requires that the path to this package is in python search path)::
$ python -m bbob_pproc.rungeneric1 -h
This will print out this help message.
* From the python interpreter (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.rungeneric1.main('-o outputfolder folder1'.split())
This will execute the post-processing on the index files found in
:file:`folder1`. The ``-o`` option changes the output folder from
the default to :file:`outputfolder`.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
if 1 < 3:
opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
if 11 < 3:
try:
opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
except getopt.error, msg:
raise Usage(msg)
if not (args) and not '--help' in argv and not 'h' in argv:
print 'not enough input arguments given'
print 'cave: the following options also need an argument:'
print [o for o in longoptlist if o[-1] == '=']
print 'options given:'
print opts
print 'try --help for help'
sys.exit()
inputCrE = 0.
isfigure = True
istab = True
isrldistr = True
islogloss = True
isPostProcessed = False
isPickled = False
verbose = False
outputdir = 'ppdata'
isNoisy = False
isNoiseFree = False
inputsettings = 'color'
isConv = False
isRLbased = None # allows automatic choice
isExpensive = None
# Process options
for o, a in opts:
if o in ("-v", "--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-p", "--pickle"):
isPickled = True
elif o in ("-o", "--output-dir"):
outputdir = a
elif o == "--noisy":
isNoisy = True
elif o == "--noise-free":
isNoiseFree = True
# The next 4 are for testing purpose
elif o == "--tab-only":
isfigure = False
isrldistr = False
islogloss = False
elif o == "--fig-only":
istab = False
isrldistr = False
islogloss = False
elif o == "--rld-only":
istab = False
isfigure = False
islogloss = False
elif o == "--los-only":
istab = False
isfigure = False
isrldistr = False
elif o == "--crafting-effort":
try:
inputCrE = float(a)
except ValueError:
raise Usage('Expect a valid float for flag crafting-effort.')
elif o == "--settings":
inputsettings = a
elif o == "--conv":
isConv = True
elif o == "--runlength-based":
isRLbased = True
elif o == "--expensive":
isExpensive = True # comprises runlength-based
elif o == "--not-expensive":
isExpensive = False
else:
assert False, "unhandled option"
# from bbob_pproc import bbob2010 as inset # input settings
if inputsettings == "color":
from bbob_pproc import genericsettings as inset # input settings
elif inputsettings == "grayscale":
from bbob_pproc import grayscalesettings as inset # input settings
elif inputsettings == "black-white":
from bbob_pproc import bwsettings as inset # input settings
else:
txt = ('Settings: %s is not an appropriate ' % inputsettings
+ 'argument for input flag "--settings".')
raise Usage(txt)
if 11 < 3:
from bbob_pproc import config # input settings
config.config()
import imp
# import testbedsettings as testbedsettings # input settings
try:
fp, pathname, description = imp.find_module("testbedsettings")
testbedsettings = imp.load_module("testbedsettings", fp, pathname, description)
finally:
fp.close()
if (not verbose):
warnings.simplefilter('module')
# warnings.simplefilter('ignore')
print ("Post-processing (1): will generate output " +
"data in folder %s" % outputdir)
print " this might take several minutes."
filelist = list()
for i in args:
i = i.strip()
if os.path.isdir(i):
filelist.extend(findfiles.main(i, verbose))
elif os.path.isfile(i):
filelist.append(i)
else:
txt = 'Input file or folder %s could not be found.' % i
print txt
raise Usage(txt)
dsList = DataSetList(filelist, verbose)
if not dsList:
raise Usage("Nothing to do: post-processing stopped.")
if isNoisy and not isNoiseFree:
dsList = dsList.dictByNoise().get('nzall', DataSetList())
if isNoiseFree and not isNoisy:
dsList = dsList.dictByNoise().get('noiselessall', DataSetList())
# compute maxfuneval values
dict_max_fun_evals = {}
for ds in dsList:
dict_max_fun_evals[ds.dim] = np.max((dict_max_fun_evals.setdefault(ds.dim, 0), float(np.max(ds.maxevals))))
if isRLbased is not None:
genericsettings.runlength_based_targets = isRLbased
from bbob_pproc import config
config.target_values(isExpensive, dict_max_fun_evals)
config.config()
if (verbose):
for i in dsList:
if (dict((j, i.instancenumbers.count(j)) for j in set(i.instancenumbers)) !=
inset.instancesOfInterest):
warnings.warn('The data of %s do not list ' % (i) +
'the correct instances ' +
'of function F%d.' % (i.funcId))
dictAlg = dsList.dictByAlg()
if len(dictAlg) > 1:
warnings.warn('Data with multiple algId %s ' % (dictAlg) +
'will be processed together.')
# TODO: in this case, all is well as long as for a given problem
# (given dimension and function) there is a single instance of
# DataSet associated. If there are more than one, the first one only
# will be considered... which is probably not what one would expect.
# TODO: put some errors where this case would be a problem.
# raise Usage?
if isfigure or istab or isrldistr or islogloss:
if not os.path.exists(outputdir):
os.makedirs(outputdir)
if verbose:
print 'Folder %s was created.' % (outputdir)
if isPickled:
dsList.pickle(verbose=verbose)
if isConv:
ppconverrorbars.main(dictAlg, outputdir, verbose)
if isfigure:
print "Scaling figures...",
sys.stdout.flush()
# ERT/dim vs dim.
plt.rc("axes", **inset.rcaxeslarger)
plt.rc("xtick", **inset.rcticklarger)
plt.rc("ytick", **inset.rcticklarger)
plt.rc("font", **inset.rcfontlarger)
plt.rc("legend", **inset.rclegendlarger)
ppfigdim.main(dsList, ppfigdim.values_of_interest, outputdir, verbose)
plt.rcdefaults()
print_done()
plt.rc("axes", **inset.rcaxes)
plt.rc("xtick", **inset.rctick)
plt.rc("ytick", **inset.rctick)
plt.rc("font", **inset.rcfont)
plt.rc("legend", **inset.rclegend)
if istab:
print "TeX tables...",
sys.stdout.flush()
dictNoise = dsList.dictByNoise()
for noise, sliceNoise in dictNoise.iteritems():
pptable.main(sliceNoise, inset.tabDimsOfInterest, outputdir,
noise, verbose)
print_done()
if isrldistr:
print "ECDF graphs...",
sys.stdout.flush()
dictNoise = dsList.dictByNoise()
if len(dictNoise) > 1:
warnings.warn('Data for functions from both the noisy and '
'non-noisy testbeds have been found. Their '
'results will be mixed in the "all functions" '
'ECDF figures.')
dictDim = dsList.dictByDim()
for dim in inset.rldDimsOfInterest:
try:
sliceDim = dictDim[dim]
except KeyError:
continue
pprldistr.main(sliceDim, True,
outputdir, 'all', verbose)
dictNoise = sliceDim.dictByNoise()
for noise, sliceNoise in dictNoise.iteritems():
pprldistr.main(sliceNoise, True,
outputdir, '%s' % noise, verbose)
dictFG = sliceDim.dictByFuncGroup()
for fGroup, sliceFuncGroup in dictFG.items():
pprldistr.main(sliceFuncGroup,
True, outputdir, '%s' % fGroup, verbose)
pprldistr.fmax = None # Resetting the max final value
pprldistr.evalfmax = None # Resetting the max #fevalsfactor
print_done()
if islogloss:
print "ERT loss ratio figures and tables...",
sys.stdout.flush()
for ng, sliceNoise in dsList.dictByNoise().iteritems():
if ng == 'noiselessall':
testbed = 'noiseless'
elif ng == 'nzall':
testbed = 'noisy'
txt = ("Please input crafting effort value "
+ "for %s testbed:\n CrE = " % testbed)
CrE = inputCrE
while CrE is None:
try:
CrE = float(raw_input(txt))
except (SyntaxError, NameError, ValueError):
print "Float value required."
dictDim = sliceNoise.dictByDim()
for d in inset.rldDimsOfInterest:
try:
sliceDim = dictDim[d]
except KeyError:
continue
info = '%s' % ng
pplogloss.main(sliceDim, CrE, True, outputdir, info,
verbose=verbose)
pplogloss.generateTable(sliceDim, CrE, outputdir, info,
verbose=verbose)
for fGroup, sliceFuncGroup in sliceDim.dictByFuncGroup().iteritems():
info = '%s' % fGroup
pplogloss.main(sliceFuncGroup, CrE, True, outputdir, info,
verbose=verbose)
pplogloss.evalfmax = None # Resetting the max #fevalsfactor
print_done()
latex_commands_file = os.path.join(outputdir.split(os.sep)[0], 'bbob_pproc_commands.tex')
prepend_to_file(latex_commands_file,
['\\providecommand{\\bbobloglosstablecaption}[1]{',
pplogloss.table_caption, '}'])
prepend_to_file(latex_commands_file,
['\\providecommand{\\bbobloglossfigurecaption}[1]{',
pplogloss.figure_caption, '}'])
prepend_to_file(latex_commands_file,
['\\providecommand{\\bbobpprldistrlegend}[1]{',
pprldistr.caption_single(np.max([ val / dim for dim, val in dict_max_fun_evals.iteritems()])), # depends on the config setting, should depend on maxfevals
'}'])
prepend_to_file(latex_commands_file,
['\\providecommand{\\bbobppfigdimlegend}[1]{',
ppfigdim.scaling_figure_caption(),
'}'])
prepend_to_file(latex_commands_file,
['\\providecommand{\\bbobpptablecaption}[1]{',
pptable.table_caption,
'}'])
prepend_to_file(latex_commands_file,
['\\providecommand{\\algfolder}{}']) # is overwritten in rungeneric.py
prepend_to_file(latex_commands_file,
['\\providecommand{\\algname}{' +
(str_to_latex(strip_pathname(args[0])) if len(args) == 1 else str_to_latex(dsList[0].algId)) + '{}}'])
if isfigure or istab or isrldistr or islogloss:
print "Output data written to folder %s" % outputdir
plt.rcdefaults()
def main(argv=None):
"""Generate python-formatted data from raw BBOB experimental data.
The raw experimental data (files with the extension 'info' pointing to
files with extension 'dat' and 'tdat') are post-processed and stored in a
more condensed way as files with the extension 'pickle'. Supposing the
raw data are stored in folder 'mydata', the new pickle files will be put in
folder 'mydata-pickle'.
Running this will also add an entry in file algorithmshortinfos.txt if it
does not exist already.
algorithmshortinfos.txt is a file which contain meta-information that are
used by modules from the bbob_pproc.compall package.
The new entry in algorithmshortinfos.txt is represented as a new line
appended at the end of the file.
The line in question will have 3 fields separated by colon (:) character.
The 1st field must be the exact string used as algId in the info files in
your data, the 2nd the exact string for the comment. The 3rd will be
a python dictionary which will be used for the plotting.
Keyword arguments:
argv -- list of strings containing options and arguments. If not provided,
sys.argv is accessed.
argv should list either names of info files or folders containing info
files or folders containing pickle files (preferred).
Furthermore, argv can begin with, in any order, facultative option
flags listed below.
-h, --help
display this message
-v, --verbose
verbose mode, prints out operations. When not in verbose mode, no
output is to be expected, except for errors.
Exceptions raised:
Usage -- Gives back a usage message.
Examples:
* Calling the dataoutput.py interface from the command line:
$ python bbob_pproc/dataoutput.py -v
$ python bbob_pproc/dataoutput.py experiment2/*.info
* Loading this package and calling the main from the command line
(requires that the path to this package is in python search path):
$ python -m bbob_pproc.dataoutput -h
This will print out this help message.
* From the python interactive shell (requires that the path to this
package is in python search path):
>>> from bbob_pproc import dataoutput
>>> dataoutput.main('folder1')
"""
if argv is None:
argv = sys.argv[1:]
try:
try:
opts, args = getopt.getopt(argv, "hv", ["help", "verbose"])
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
verbose = False
#Process options
for o, a in opts:
if o in ("-v", "--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
else:
assert False, "unhandled option"
if (not verbose):
warnings.simplefilter('ignore')
dsList = DataSetList(args)
outputPickle(dsList, verbose=verbose)
sys.exit()
def main(argv=None):
"""Main routine for post-processing the data of multiple algorithms.
Keyword arguments:
argv -- list of strings containing options and arguments. If not provided,
sys.argv is accessed.
argv must list folders containing BBOB data files. Each of these folders
should correspond to the data of ONE algorithm and should be listed in
algorithmshortinfos.txt, a file from the bbob_pproc.compall package listing
the information of various algorithms treated using bbob_pproc.dataoutput
Furthermore, argv can begin with, in any order, facultative option flags
listed below.
-h, --help
display this message
-v, --verbose
verbose mode, prints out operations. When not in verbose mode, no
output is to be expected, except for errors.
-o, --output-dir OUTPUTDIR
change the default output directory ('defaultoutputdirectory') to
OUTPUTDIR
--noise-free, --noisy
restrain the post-processing to part of the data set only. Actually
quicken the post-processing since it loads only part of the pickle
files.
--tab-only, --perfprof-only
these options can be used to output respectively the comparison
tex tables or the performance profiles only.
A combination of any two of these options results in
no output.
Exceptions raised:
Usage -- Gives back a usage message.
Examples:
* Calling the runcompall.py interface from the command line:
$ python bbob_pproc/runcompall.py -v
* Loading this package and calling the main from the command line
(requires that the path to this package is in python search path):
$ python -m bbob_pproc.runcompall -h
This will print out this help message.
* From the python interactive shell (requires that the path to this
package is in python search path):
>>> from bbob_pproc import runcompall
>>> runcompall.main('-o outputfolder folder1 folder2'.split())
This will execute the post-processing on the data found in folder1
and folder2.
The -o option changes the output folder from the default cmpalldata to
outputfolder.
* Generate post-processing data for some algorithms:
$ python runcompall.py AMALGAM BFGS CMA-ES
"""
if argv is None:
argv = sys.argv[1:]
try:
try:
opts, args = getopt.getopt(argv, "hvo:",
["help", "output-dir=", "noisy",
"noise-free", "perfprof-only",
"tab-only", "verbose"])
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
verbose = False
outputdir = 'cmpalldata'
isNoisy = False
isNoiseFree = False
isPer = True
isTab = True
#Process options
for o, a in opts:
if o in ("-v","--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-o", "--output-dir"):
outputdir = a
elif o == "--noisy":
isNoisy = True
elif o == "--noise-free":
isNoiseFree = True
elif o == "--tab-only":
isPer = False
isEff = False
elif o == "--perfprof-only":
isEff = False
isTab = False
else:
assert False, "unhandled option"
if (not verbose):
warnings.simplefilter('ignore')
print ("BBOB Post-processing: will generate comparison " +
"data in folder %s" % outputdir)
print " this might take several minutes."
dsList, sortedAlgs, dictAlg = processInputArgs(args, verbose=verbose)
if not dsList:
sys.exit()
for i in dictAlg:
if isNoisy and not isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get('nzall', DataSetList())
elif isNoiseFree and not isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get('noiselessall', DataSetList())
tmp = set((j.algId, j.comment) for j in dictAlg[i])
for j in tmp:
if not dataoutput.isListed(j):
dataoutput.updateAlgorithmInfo(j, verbose=verbose)
for i in dsList:
if not i.dim in (2, 3, 5, 10, 20):
continue
# Deterministic algorithms
if i.algId in ('Original DIRECT', ):
tmpInstancesOfInterest = instancesOfInterestDet
else:
tmpInstancesOfInterest = instancesOfInterest
if ((dict((j, i.itrials.count(j)) for j in set(i.itrials)) <
tmpInstancesOfInterest) and
(dict((j, i.itrials.count(j)) for j in set(i.itrials)) <
instancesOfInterest2010)):
warnings.warn('The data of %s do not list ' %(i) +
'the correct instances ' +
'of function F%d or the ' %(i.funcId) +
'correct number of trials for each.')
# group targets:
dictTarget = {}
for t in sorted(set(single_target_function_values + summarized_target_function_values)):
tmpdict = dict.fromkeys(((f, d) for f in range(0, 25) + range(101, 131) for d in (2, 3, 5, 10, 20, 40)), t)
stmp = 'E'
if t == 1:
stmp = 'E-'
# dictTarget['_f' + stmp + '%2.1f' % numpy.log10(t)] = (tmpdict, )
if t in single_target_function_values:
dictTarget['_f' + stmp + '%02d' % numpy.log10(t)] = (tmpdict, )
if t in summarized_target_function_values:
dictTarget.setdefault('_allfs', []).append(tmpdict)
if not os.path.exists(outputdir):
os.mkdir(outputdir)
if verbose:
print 'Folder %s was created.' % (outputdir)
# Performance profiles
if isPer:
dictNoi = pproc.dictAlgByNoi(dictAlg)
for ng, tmpdictAlg in dictNoi.iteritems():
dictDim = pproc.dictAlgByDim(tmpdictAlg)
for d, entries in dictDim.iteritems():
for k, t in dictTarget.iteritems():
#set_trace()
ppperfprof.main(entries, target=t, order=sortedAlgs,
plotArgs=algPlotInfos,
outputdir=outputdir,
info=('%02d%s_%s' % (d, k, ng)),
verbose=verbose)
organizeRTDpictures.do(outputdir)
print "ECDFs of ERT figures done."
if isTab:
allmintarget, allertbest = detTarget(dsList)
pptables.tablemanyalgonefunc(dictAlg, allmintarget, allertbest,
sortedAlgs, outputdir)
print "Comparison tables done."
def main(argv=None):
"""Generates from BBOB experiment data some outputs for a tex document.
Provided with some index entries (found in files with the 'info' extension)
this routine outputs figure and TeX files in the folder 'ppdata' needed for
the compilation of latex document templateBBOBarticle.tex. These output
files will contain performance tables, performance scaling figures and
empirical cumulative distribution figures. On subsequent executions, new
files will be added to the output directory, overwriting existing older
files in the process.
Keyword arguments:
argv -- list of strings containing options and arguments. If not given,
sys.argv is accessed.
argv should list either names of info files or folders containing info
files. argv can also contain post-processed pickle files generated by this
routine. Furthermore, argv can begin with, in any order, facultative option
flags listed below.
-h, --help
display this message
-v, --verbose
verbose mode, prints out operations. When not in verbose mode, no
output is to be expected, except for errors.
-p, --pickle
generates pickle post processed data files.
-o, --output-dir OUTPUTDIR
change the default output directory ('ppdata') to OUTPUTDIR
--crafting-effort=VALUE
sets the crafting effort to VALUE. Otherwise the user will be
prompted. This flag is useful when running this script in batch.
-f, --final
lengthens the bootstrapping process used as dispersion measure in
the tables generation. This might at least double the time of the
whole post-processing. Please use this option when generating your
final paper.
--tab-only, --fig-only, --rld-only, --los-only
these options can be used to output respectively the tex tables,
convergence and ENFEs graphs figures, run length distribution
figures, ERT loss ratio figures only. A combination of any two of
these options results in no output.
Exceptions raised:
Usage -- Gives back a usage message.
Examples:
* Calling the run.py interface from the command line:
$ python bbob_pproc/run.py -v experiment1
will post-process the folder experiment1 and all its containing data,
base on the found .info files in the folder. The result will appear
in folder ppdata. The -v option adds verbosity.
$ python bbob_pproc/run.py -o otherppdata experiment2/*.info
This will execute the post-processing on the info files found in
experiment2. The result will be located in the alternative location
otherppdata.
* Loading this package and calling the main from the command line
(requires that the path to this package is in python search path):
$ python -m bbob_pproc -h
This will print out this help message.
* From the python interactive shell (requires that the path to this
package is in python search path):
>>> import bbob_pproc
>>> bbob_pproc.main('-o outputfolder folder1'.split())
This will execute the post-processing on the index files found in folder1.
The -o option changes the output folder from the default ppdata to
outputfolder.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
try:
try:
opts, args = getopt.getopt(argv, "hvpfo:", [
"help", "output-dir=", "tab-only", "fig-only", "rld-only",
"los-only", "crafting-effort=", "pickle", "verbose", "final"
])
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
CrE = None
isfigure = True
istab = True
isrldistr = True
islogloss = True
isPostProcessed = False
isPickled = False
isDraft = True
verbose = False
outputdir = 'ppdata'
#Process options
for o, a in opts:
if o in ("-v", "--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-p", "--pickle"):
isPickled = True
elif o in ("-o", "--output-dir"):
outputdir = a
elif o in ("-f", "--final"):
isDraft = False
#The next 3 are for testing purpose
elif o == "--tab-only":
isfigure = False
isrldistr = False
islogloss = False
elif o == "--fig-only":
istab = False
isrldistr = False
islogloss = False
elif o == "--rld-only":
istab = False
isfigure = False
islogloss = False
elif o == "--los-only":
istab = False
isfigure = False
isrldistr = False
elif o == "--crafting-effort":
try:
CrE = float(a)
except ValueError:
raise Usage(
'Expect a valid float for flag crafting-effort.')
else:
assert False, "unhandled option"
if (not verbose):
warnings.simplefilter('ignore')
print("BBOB Post-processing: will generate post-processing " +
"data in folder %s" % outputdir)
print " this might take several minutes."
filelist = list()
for i in args:
if os.path.isdir(i):
filelist.extend(findfiles.main(i, verbose))
elif os.path.isfile(i):
filelist.append(i)
else:
txt = 'Input file or folder %s could not be found.'
raise Usage(txt)
dsList = DataSetList(filelist, verbose)
if not dsList:
raise Usage("Nothing to do: post-processing stopped.")
if (verbose):
for i in dsList:
if (dict((j, i.itrials.count(j))
for j in set(i.itrials)) != instancesOfInterest2010):
warnings.warn('The data of %s do not list ' % (i) +
'the correct instances ' +
'of function F%d.' % (i.funcId))
# BBOB 2009 Checking
#if ((dict((j, i.itrials.count(j)) for j in set(i.itrials)) !=
#instancesOfInterest) and
#(dict((j, i.itrials.count(j)) for j in set(i.itrials)) !=
#instancesOfInterest2010)):
#warnings.warn('The data of %s do not list ' %(i) +
#'the correct instances ' +
#'of function F%d or the ' %(i.funcId) +
#'correct number of trials for each.')
dictAlg = dsList.dictByAlg()
if len(dictAlg) > 1:
warnings.warn('Data with multiple algId %s ' % (dictAlg) +
'will be processed together.')
#TODO: in this case, all is well as long as for a given problem
#(given dimension and function) there is a single instance of
#DataSet associated. If there are more than one, the first one only
#will be considered... which is probably not what one would expect.
#TODO: put some errors where this case would be a problem.
if isfigure or istab or isrldistr or islogloss:
if not os.path.exists(outputdir):
os.mkdir(outputdir)
if verbose:
print 'Folder %s was created.' % (outputdir)
if isPickled:
dsList.pickle(verbose=verbose)
if isfigure:
ppfigdim.main(dsList, figValsOfInterest, outputdir, verbose)
#ERT/dim vs dim.
#ppfigdim.main2(dsList, figValsOfInterest, outputdir,
#verbose)
print "Scaling figures done."
if istab:
dictFunc = dsList.dictByFunc()
for fun, sliceFun in dictFunc.items():
dictDim = sliceFun.dictByDim()
tmp = []
for dim in tabDimsOfInterest:
try:
if len(dictDim[dim]) > 1:
warnings.warn(
'Func: %d, DIM %d: ' % (fun, dim) +
'multiple index entries. Will only ' +
'process the first ' + '%s.' % dictDim[dim][0])
tmp.append(dictDim[dim][0])
except KeyError:
pass
if tmp:
filename = os.path.join(outputdir, 'ppdata_f%d' % fun)
pptex.main(tmp, tabValsOfInterest, filename, isDraft,
verbose)
print "TeX tables",
if isDraft:
print(
"(draft) done. To get final version tables, please "
"use the -f option with run.py")
else:
print "done."
if isrldistr:
dictNoise = dsList.dictByNoise()
if len(dictNoise) > 1:
warnings.warn('Data for functions from both the noisy and '
'non-noisy testbeds have been found. Their '
'results will be mixed in the "all functions" '
'ECDF figures.')
dictDim = dsList.dictByDim()
for dim in rldDimsOfInterest:
try:
sliceDim = dictDim[dim]
pprldistr.main(sliceDim, rldValsOfInterest, True,
outputdir, 'dim%02dall' % dim, verbose)
dictNoise = sliceDim.dictByNoise()
for noise, sliceNoise in dictNoise.iteritems():
pprldistr.main(sliceNoise, rldValsOfInterest, True,
outputdir, 'dim%02d%s' % (dim, noise),
verbose)
dictFG = sliceDim.dictByFuncGroup()
for fGroup, sliceFuncGroup in dictFG.items():
pprldistr.main(sliceFuncGroup, rldValsOfInterest, True,
outputdir, 'dim%02d%s' % (dim, fGroup),
verbose)
pprldistr.fmax = None #Resetting the max final value
pprldistr.evalfmax = None #Resetting the max #fevalsfactor
except KeyError:
pass
print "ECDF graphs done."
if islogloss:
for ng, sliceNoise in dsList.dictByNoise().iteritems():
if ng == 'noiselessall':
testbed = 'noiseless'
elif ng == 'nzall':
testbed = 'noisy'
txt = ("Please input crafting effort value " +
"for %s testbed:\n CrE = " % testbed)
while CrE is None:
try:
CrE = float(input(txt))
except (SyntaxError, NameError, ValueError):
print "Float value required."
dictDim = sliceNoise.dictByDim()
for d in rldDimsOfInterest:
try:
sliceDim = dictDim[d]
except KeyError:
continue
info = 'dim%02d%s' % (d, ng)
pplogloss.main(sliceDim,
CrE,
True,
outputdir,
info,
verbose=verbose)
pplogloss.generateTable(sliceDim,
CrE,
outputdir,
info,
verbose=verbose)
for fGroup, sliceFuncGroup in sliceDim.dictByFuncGroup(
).iteritems():
info = 'dim%02d%s' % (d, fGroup)
pplogloss.main(sliceFuncGroup,
CrE,
True,
outputdir,
info,
verbose=verbose)
pplogloss.evalfmax = None #Resetting the max #fevalsfactor
print "ERT loss ratio figures and tables done."
if isfigure or istab or isrldistr or islogloss:
print "Output data written to folder %s." % outputdir
def main(argv=None):
r"""Routine for post-processing COCO data from two algorithms.
Provided with some data, this routine outputs figure and TeX files
in a folder needed for the compilation of the provided LaTeX templates
for comparing two algorithms (``*cmp.tex`` or ``*2*.tex``).
The used template file needs to be edited so that the command
``\bbobdatapath`` points to the output folder created by this routine.
The output files will contain performance tables, performance
scaling figures and empirical cumulative distribution figures. On
subsequent executions, new files will be added to the output folder,
overwriting existing older files in the process.
Keyword arguments:
*argv* -- list of strings containing options and arguments. If not
given, sys.argv is accessed.
*argv* must list folders containing BBOB data files. Each of these
folders should correspond to the data of ONE algorithm.
Furthermore, argv can begin with, in any order, facultative option
flags listed below.
-h, --help
displays this message.
-v, --verbose
verbose mode, prints out operations.
-o OUTPUTDIR, --output-dir=OUTPUTDIR
changes the default output directory (:file:`ppdata`) to
:file:`OUTPUTDIR`
--noise-free, --noisy
processes only part of the data.
--settings=SETTING
changes the style of the output figures and tables. At the
moment only the only differences are in the colors of the
output figures. SETTING can be either "grayscale", "color"
or "black-white". The default setting is "color".
--fig-only, --rld-only, --tab-only, --sca-only
these options can be used to output respectively the ERT
graphs figures, run length distribution figures or the
comparison tables scatter plot figures only. Any combination
of these options results in no output.
--conv
if this option is chosen, additionally convergence
plots for each function and algorithm are generated.
--rld-single-fcts
generate also runlength distribution figures for each
single function.
--expensive
runlength-based f-target values and fixed display limits,
useful with comparatively small budgets. By default the
setting is based on the budget used in the data.
--not-expensive
expensive setting off.
--svg
generate also the svg figures which are used in html files
Exceptions raised:
*Usage* -- Gives back a usage message.
Examples:
* Calling the rungeneric2.py interface from the command line::
$ python bbob_pproc/rungeneric2.py -v Alg0-baseline Alg1-of-interest
will post-process the data from folders :file:`Alg0-baseline` and
:file:`Alg1-of-interest`, the former containing data for the
reference algorithm (zero-th) and the latter data for the
algorithm of concern (first). The results will be output in the
default output folder. The ``-v`` option adds verbosity.
* From the python interpreter (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.rungeneric2.main('-o outputfolder PSO DEPSO'.split())
This will execute the post-processing on the data found in folder
:file:`PSO` and :file:`DEPSO`. The ``-o`` option changes the output
folder from the default to :file:`outputfolder`.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
global ftarget
try:
try:
opts, args = getopt.getopt(argv, genericsettings.shortoptlist,
genericsettings.longoptlist)
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
#Process options
outputdir = genericsettings.outputdir
for o, a in opts:
if o in ("-v", "--verbose"):
genericsettings.verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-o", "--output-dir"):
outputdir = a
elif o == "--fig-only":
genericsettings.isRLDistr = False
genericsettings.isTab = False
genericsettings.isScatter = False
elif o == "--rld-only":
genericsettings.isFig = False
genericsettings.isTab = False
genericsettings.isScatter = False
elif o == "--tab-only":
genericsettings.isFig = False
genericsettings.isRLDistr = False
genericsettings.isScatter = False
elif o == "--sca-only":
genericsettings.isFig = False
genericsettings.isRLDistr = False
genericsettings.isTab = False
elif o == "--noisy":
genericsettings.isNoisy = True
elif o == "--noise-free":
genericsettings.isNoiseFree = True
elif o == "--settings":
genericsettings.inputsettings = a
elif o == "--conv":
genericsettings.isConv = True
elif o == "--rld-single-fcts":
genericsettings.isRldOnSingleFcts = True
elif o == "--runlength-based":
genericsettings.runlength_based_targets = True
elif o == "--expensive":
genericsettings.isExpensive = True # comprises runlength-based
elif o == "--not-expensive":
genericsettings.isExpensive = False
elif o == "--svg":
genericsettings.generate_svg_files = True
elif o == "--los-only":
warnings.warn(
"option --los-only will have no effect with rungeneric2.py"
)
elif o == "--crafting-effort=":
warnings.warn(
"option --crafting-effort will have no effect with rungeneric2.py"
)
elif o in ("-p", "--pickle"):
warnings.warn(
"option --pickle will have no effect with rungeneric2.py")
else:
assert False, "unhandled option"
# from bbob_pproc import bbob2010 as inset # input settings
if genericsettings.inputsettings == "color":
from bbob_pproc import genericsettings as inset # input settings
config.config()
elif genericsettings.inputsettings == "grayscale": # probably very much obsolete
from bbob_pproc import grayscalesettings as inset # input settings
elif genericsettings.inputsettings == "black-white": # probably very much obsolete
from bbob_pproc import bwsettings as inset # input settings
else:
txt = ('Settings: %s is not an appropriate ' %
genericsettings.inputsettings +
'argument for input flag "--settings".')
raise Usage(txt)
if (not genericsettings.verbose):
warnings.simplefilter('module')
warnings.simplefilter('ignore')
print("Post-processing will generate comparison " +
"data in folder %s" % outputdir)
print " this might take several minutes."
dsList, sortedAlgs, dictAlg = processInputArgs(
args, verbose=genericsettings.verbose)
if 1 < 3 and len(sortedAlgs) != 2:
raise ValueError(
'rungeneric2.py needs exactly two algorithms to compare, found: '
+ str(sortedAlgs) +
'\n use rungeneric.py (or rungenericmany.py) to compare more algorithms. '
)
if not dsList:
sys.exit()
for i in dictAlg:
if genericsettings.isNoisy and not genericsettings.isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'nzall', DataSetList())
if genericsettings.isNoiseFree and not genericsettings.isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'noiselessall', DataSetList())
for i in dsList:
if i.dim not in genericsettings.dimensions_to_display:
continue
if (dict((j, i.instancenumbers.count(j))
for j in set(i.instancenumbers)) <
inset.instancesOfInterest):
warnings.warn('The data of %s do not list ' % (i) +
'the correct instances ' + 'of function F%d.' %
(i.funcId))
if len(sortedAlgs) < 2:
raise Usage('Expect data from two different algorithms, could ' +
'only find one.')
elif len(sortedAlgs) > 2:
warnings.warn('Data from folders: %s ' % (sortedAlgs) +
'were found, the first two will be processed.')
# Group by algorithm
dsList0 = dictAlg[sortedAlgs[0]]
if not dsList0:
raise Usage('Could not find data for algorithm %s.' %
(sortedAlgs[0]))
dsList1 = dictAlg[sortedAlgs[1]]
if not dsList1:
raise Usage('Could not find data for algorithm %s.' %
(sortedAlgs[0]))
# get the name of each algorithm from the input arguments
tmppath0, alg0name = os.path.split(sortedAlgs[0].rstrip(os.sep))
tmppath1, alg1name = os.path.split(sortedAlgs[1].rstrip(os.sep))
for i in dsList0:
i.algId = alg0name
for i in dsList1:
i.algId = alg1name
# compute maxfuneval values
dict_max_fun_evals1 = {}
dict_max_fun_evals2 = {}
for ds in dsList0:
dict_max_fun_evals1[ds.dim] = np.max(
(dict_max_fun_evals1.setdefault(ds.dim, 0),
float(np.max(ds.maxevals))))
for ds in dsList1:
dict_max_fun_evals2[ds.dim] = np.max(
(dict_max_fun_evals2.setdefault(ds.dim, 0),
float(np.max(ds.maxevals))))
config.target_values(
genericsettings.isExpensive, {
1:
min([
max([
val / dim
for dim, val in dict_max_fun_evals1.iteritems()
]),
max([
val / dim
for dim, val in dict_max_fun_evals2.iteritems()
])
])
})
config.config()
######################### Post-processing #############################
if genericsettings.isFig or genericsettings.isRLDistr or genericsettings.isTab or genericsettings.isScatter:
if not os.path.exists(outputdir):
os.mkdir(outputdir)
if genericsettings.verbose:
print 'Folder %s was created.' % (outputdir)
# prepend the algorithm name command to the tex-command file
abc = 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'
lines = []
for i, alg in enumerate(args):
lines.append('\\providecommand{\\algorithm' + abc[i] + '}{' +
str_to_latex(strip_pathname1(alg)) + '}')
prepend_to_file(
os.path.join(outputdir,
'bbob_pproc_commands.tex'), lines, 1000,
'bbob_proc_commands.tex truncated, consider removing the file before the text run'
)
# Check whether both input arguments list noisy and noise-free data
dictFN0 = dsList0.dictByNoise()
dictFN1 = dsList1.dictByNoise()
k0 = set(dictFN0.keys())
k1 = set(dictFN1.keys())
symdiff = k1 ^ k0 # symmetric difference
if symdiff:
tmpdict = {}
for i, noisegrp in enumerate(symdiff):
if noisegrp == 'nzall':
tmp = 'noisy'
elif noisegrp == 'noiselessall':
tmp = 'noiseless'
if dictFN0.has_key(noisegrp):
tmp2 = sortedAlgs[0]
elif dictFN1.has_key(noisegrp):
tmp2 = sortedAlgs[1]
tmpdict.setdefault(tmp2, []).append(tmp)
txt = []
for i, j in tmpdict.iteritems():
txt.append('Only input folder %s lists %s data.' %
(i, ' and '.join(j)))
raise Usage('Data Mismatch: \n ' + ' '.join(txt) +
'\nTry using --noise-free or --noisy flags.')
if genericsettings.isFig:
plt.rc("axes", **inset.rcaxeslarger)
plt.rc("xtick", **inset.rcticklarger)
plt.rc("ytick", **inset.rcticklarger)
plt.rc("font", **inset.rcfontlarger)
plt.rc("legend", **inset.rclegendlarger)
plt.rc('pdf', fonttype=42)
ppfig2.main(dsList0, dsList1, ppfig2_ftarget, outputdir,
genericsettings.verbose)
print "log ERT1/ERT0 vs target function values done."
plt.rc("axes", **inset.rcaxes)
plt.rc("xtick", **inset.rctick)
plt.rc("ytick", **inset.rctick)
plt.rc("font", **inset.rcfont)
plt.rc("legend", **inset.rclegend)
plt.rc('pdf', fonttype=42)
if genericsettings.isRLDistr:
if len(dictFN0) > 1 or len(dictFN1) > 1:
warnings.warn('Data for functions from both the noisy and ' +
'non-noisy testbeds have been found. Their ' +
'results will be mixed in the "all functions" ' +
'ECDF figures.')
dictDim0 = dsList0.dictByDim()
dictDim1 = dsList1.dictByDim()
# ECDFs of ERT ratios
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
if dim in inset.rldDimsOfInterest:
# ECDF for all functions altogether
try:
pprldistr2.main(dictDim0[dim], dictDim1[dim], dim,
inset.rldValsOfInterest, outputdir,
'%02dD_all' % dim,
genericsettings.verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.' %
(dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr2.main(dictFG1[fGroup], dictFG0[fGroup], dim,
inset.rldValsOfInterest, outputdir,
'%02dD_%s' % (dim, fGroup),
genericsettings.verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr2.main(dictFN1[fGroup], dictFN0[fGroup], dim,
inset.rldValsOfInterest, outputdir,
'%02dD_%s' % (dim, fGroup),
genericsettings.verbose)
prepend_to_file(
os.path.join(outputdir, 'bbob_pproc_commands.tex'),
[
'\\providecommand{\\bbobpprldistrlegendtwo}[1]{',
pprldistr.caption_two(
), # depends on the config setting, should depend on maxfevals
'}'
])
print "ECDF runlength ratio graphs done."
for dim in set(dictDim0.keys()) & set(dictDim1.keys()):
pprldistr.fmax = None #Resetting the max final value
pprldistr.evalfmax = None #Resetting the max #fevalsfactor
# ECDFs of all functions altogether
if dim in inset.rldDimsOfInterest:
try:
pprldistr.comp(
dictDim1[dim],
dictDim0[dim],
inset.
rldValsOfInterest, # TODO: let rldVals... possibly be RL-based targets
True,
outputdir,
'all',
genericsettings.verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.' %
(dim))
continue
# ECDFs per function groups
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) & set(dictFG1.keys()):
pprldistr.comp(dictFG1[fGroup], dictFG0[fGroup],
inset.rldValsOfInterest, True,
outputdir, '%s' % fGroup,
genericsettings.verbose)
# ECDFs per noise groups
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pprldistr.comp(dictFN1[fGroup], dictFN0[fGroup],
inset.rldValsOfInterest, True,
outputdir, '%s' % fGroup,
genericsettings.verbose)
if genericsettings.isRldOnSingleFcts: # copy-paste from above, here for each function instead of function groups
# ECDFs for each function
pprldmany.all_single_functions(dictAlg, sortedAlgs, outputdir,
genericsettings.verbose)
print "ECDF runlength graphs done."
if genericsettings.isConv:
ppconverrorbars.main(dictAlg, outputdir, genericsettings.verbose)
if genericsettings.isScatter:
if genericsettings.runlength_based_targets:
ppscatter.targets = ppscatter.runlength_based_targets
ppscatter.main(dsList1,
dsList0,
outputdir,
verbose=genericsettings.verbose)
prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'),
[
'\\providecommand{\\bbobppscatterlegend}[1]{',
ppscatter.figure_caption(), '}'
])
replace_in_file(
os.path.join(outputdir,
genericsettings.two_algorithm_file_name +
'.html'), '##bbobppscatterlegend##',
ppscatter.figure_caption_html())
print "Scatter plots done."
if genericsettings.isTab:
dictNG0 = dsList0.dictByNoise()
dictNG1 = dsList1.dictByNoise()
for nGroup in set(dictNG0.keys()) & set(dictNG1.keys()):
# split table in as many as necessary
dictFunc0 = dictNG0[nGroup].dictByFunc()
dictFunc1 = dictNG1[nGroup].dictByFunc()
funcs = list(set(dictFunc0.keys()) & set(dictFunc1.keys()))
if len(funcs) > 24:
funcs.sort()
nbgroups = int(numpy.ceil(len(funcs) / 24.))
def split_seq(seq, nbgroups):
newseq = []
splitsize = 1.0 / nbgroups * len(seq)
for i in range(nbgroups):
newseq.append(
seq[int(round(i * splitsize)
):int(round((i + 1) * splitsize))])
return newseq
groups = split_seq(funcs, nbgroups)
# merge
group0 = []
group1 = []
for i, g in enumerate(groups):
tmp0 = DataSetList()
tmp1 = DataSetList()
for f in g:
tmp0.extend(dictFunc0[f])
tmp1.extend(dictFunc1[f])
group0.append(tmp0)
group1.append(tmp1)
for i, g in enumerate(zip(group0, group1)):
pptable2.main(g[0], g[1], inset.tabDimsOfInterest,
outputdir, '%s%d' % (nGroup, i),
genericsettings.verbose)
else:
if 11 < 3: # future handling:
dictFunc0 = dsList0.dictByFunc()
dictFunc1 = dsList1.dictByFunc()
funcs = list(
set(dictFunc0.keys()) & set(dictFunc1.keys()))
funcs.sort()
# nbgroups = int(numpy.ceil(len(funcs)/testbedsettings.numberOfFunctions))
# pptable2.main(dsList0, dsList1,
# testbedsettings.tabDimsOfInterest, outputdir,
# '%s' % (testbedsettings.testbedshortname), genericsettings.verbose)
else:
pptable2.main(dictNG0[nGroup], dictNG1[nGroup],
inset.tabDimsOfInterest, outputdir,
'%s' % (nGroup), genericsettings.verbose)
if isinstance(pptable2.targetsOfInterest,
pproc.RunlengthBasedTargetValues):
prepend_to_file(
os.path.join(outputdir, 'bbob_pproc_commands.tex'), [
'\\providecommand{\\bbobpptablestwolegend}[1]{',
pptable2.table_caption_expensive, '}'
])
else:
prepend_to_file(
os.path.join(outputdir, 'bbob_pproc_commands.tex'), [
'\\providecommand{\\bbobpptablestwolegend}[1]{',
pptable2.table_caption, '}'
])
htmlFileName = os.path.join(
outputdir, genericsettings.two_algorithm_file_name + '.html')
key = '##bbobpptablestwolegendexpensive##' if isinstance(
pptable2.targetsOfInterest, pproc.RunlengthBasedTargetValues
) else '##bbobpptablestwolegend##'
replace_in_file(htmlFileName, '##bbobpptablestwolegend##',
htmldesc.getValue(key))
alg0 = set(i[0] for i in dsList0.dictByAlg().keys()).pop().replace(
genericsettings.extraction_folder_prefix, '')[0:3]
alg1 = set(i[0] for i in dsList1.dictByAlg().keys()).pop().replace(
genericsettings.extraction_folder_prefix, '')[0:3]
replace_in_file(htmlFileName, 'algorithmAshort', alg0)
replace_in_file(htmlFileName, 'algorithmBshort', alg1)
for i, alg in enumerate(args):
replace_in_file(htmlFileName, 'algorithm' + abc[i],
str_to_latex(strip_pathname1(alg)))
print "Tables done."
if genericsettings.isScaleUp:
plt.rc("axes", labelsize=20, titlesize=24)
plt.rc("xtick", labelsize=20)
plt.rc("ytick", labelsize=20)
plt.rc("font", size=20)
plt.rc("legend", fontsize=20)
plt.rc('pdf', fonttype=42)
if genericsettings.runlength_based_targets:
ftarget = RunlengthBasedTargetValues([
target_runlength
]) # TODO: make this more variable but also consistent
ppfigs.main(dictAlg, genericsettings.two_algorithm_file_name,
sortedAlgs, ftarget, outputdir,
genericsettings.verbose)
plt.rcdefaults()
print "Scaling figures done."
if genericsettings.isFig or genericsettings.isRLDistr or genericsettings.isTab or genericsettings.isScatter or genericsettings.isScaleUp:
print "Output data written to folder %s" % outputdir
plt.rcdefaults()
def main(argv=None):
r"""Post-processing COCO data of a single algorithm.
Provided with some data, this routine outputs figure and TeX files
in a folder needed for the compilation of the provided LaTeX templates
for one algorithm (``*article.tex`` or ``*1*.tex``).
The used template file needs to be edited so that the commands
``\bbobdatapath`` and ``\algfolder`` point to the output folder created
by this routine.
These output files will contain performance tables, performance
scaling figures and empirical cumulative distribution figures. On
subsequent executions, new files will be added to the output folder,
overwriting existing older files in the process.
Keyword arguments:
*argv* -- list of strings containing options and arguments. If not
given, sys.argv is accessed.
*argv* should list either names of :file:`info` files or folders
containing :file:`info` files. argv can also contain post-processed
:file:`pickle` files generated by this routine. Furthermore, *argv*
can begin with, in any order, facultative option flags listed below.
-h, --help
displays this message.
-v, --verbose
verbose mode, prints out all operations.
-p, --pickle
generates pickle post processed data files.
-o OUTPUTDIR, --output-dir=OUTPUTDIR
changes the default output directory (:file:`ppdata`) to
:file:`OUTPUTDIR`.
--crafting-effort=VALUE
sets the crafting effort to VALUE (float). Otherwise the
default value of 0. will be used.
--noise-free, --noisy
processes only part of the data.
--settings=SETTINGS
changes the style of the output figures and tables. At the
moment the only differences are in the colors of the output
figures. SETTINGS can be either "grayscale", "color" or
"black-white". The default setting is "color".
--tab-only, --fig-only, --rld-only, --los-only
these options can be used to output respectively the TeX
tables, convergence and ERTs graphs figures, run length
distribution figures, ERT loss ratio figures only. A
combination of any two of these options results in no
output.
--conv
if this option is chosen, additionally convergence plots
for each function and algorithm are generated.
--rld-single-fcts
generate also runlength distribution figures for each
single function.
--expensive
runlength-based f-target values and fixed display limits,
useful with comparatively small budgets. By default the
setting is based on the budget used in the data.
--not-expensive
expensive setting off.
--svg
generate also the svg figures which are used in html files
--runlength-based
runlength-based f-target values, such that the
"level of difficulty" is similar for all functions.
Exceptions raised:
*Usage* -- Gives back a usage message.
Examples:
* Calling the rungeneric1.py interface from the command line::
$ python bbob_pproc/rungeneric1.py -v experiment1
will post-process the folder experiment1 and all its containing
data, base on the .info files found in the folder. The result will
appear in the default output folder. The -v option adds verbosity. ::
$ python bbob_pproc/rungeneric1.py -o exp2 experiment2/*.info
This will execute the post-processing on the info files found in
:file:`experiment2`. The result will be located in the alternative
location :file:`exp2`.
* Loading this package and calling the main from the command line
(requires that the path to this package is in python search path)::
$ python -m bbob_pproc.rungeneric1 -h
This will print out this help message.
* From the python interpreter (requires that the path to this
package is in python search path)::
>> import bbob_pproc as bb
>> bb.rungeneric1.main('-o outputfolder folder1'.split())
This will execute the post-processing on the index files found in
:file:`folder1`. The ``-o`` option changes the output folder from
the default to :file:`outputfolder`.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
if 1 < 3:
opts, args = getopt.getopt(argv, genericsettings.shortoptlist, genericsettings.longoptlist)
if 11 < 3:
try:
opts, args = getopt.getopt(argv, genericsettings.shortoptlist, genericsettings.longoptlist)
except getopt.error, msg:
raise Usage(msg)
if not (args) and not "--help" in argv and not "h" in argv:
print "not enough input arguments given"
print "cave: the following options also need an argument:"
print [o for o in genericsettings.longoptlist if o[-1] == "="]
print "options given:"
print opts
print "try --help for help"
sys.exit()
# Process options
outputdir = genericsettings.outputdir
for o, a in opts:
if o in ("-v", "--verbose"):
genericsettings.verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-p", "--pickle"):
genericsettings.isPickled = True
elif o in ("-o", "--output-dir"):
outputdir = a
elif o == "--noisy":
genericsettings.isNoisy = True
elif o == "--noise-free":
genericsettings.isNoiseFree = True
# The next 4 are for testing purpose
elif o == "--tab-only":
genericsettings.isFig = False
genericsettings.isRLDistr = False
genericsettings.isLogLoss = False
elif o == "--fig-only":
genericsettings.isTab = False
genericsettings.isRLDistr = False
genericsettings.isLogLoss = False
elif o == "--rld-only":
genericsettings.isTab = False
genericsettings.isFig = False
genericsettings.isLogLoss = False
elif o == "--los-only":
genericsettings.isTab = False
genericsettings.isFig = False
genericsettings.isRLDistr = False
elif o == "--crafting-effort":
try:
genericsettings.inputCrE = float(a)
except ValueError:
raise Usage("Expect a valid float for flag crafting-effort.")
elif o == "--settings":
genericsettings.inputsettings = a
elif o == "--conv":
genericsettings.isConv = True
elif o == "--rld-single-fcts":
genericsettings.isRldOnSingleFcts = True
elif o == "--runlength-based":
genericsettings.runlength_based_targets = True
elif o == "--expensive":
genericsettings.isExpensive = True # comprises runlength-based
elif o == "--not-expensive":
genericsettings.isExpensive = False
elif o == "--svg":
genericsettings.generate_svg_files = True
elif o == "--sca-only":
warnings.warn("option --sca-only will have no effect with rungeneric1.py")
else:
assert False, "unhandled option"
# from bbob_pproc import bbob2010 as inset # input settings
if genericsettings.inputsettings == "color":
from bbob_pproc import genericsettings as inset # input settings
elif genericsettings.inputsettings == "grayscale":
from bbob_pproc import grayscalesettings as inset # input settings
elif genericsettings.inputsettings == "black-white":
from bbob_pproc import bwsettings as inset # input settings
else:
txt = (
"Settings: %s is not an appropriate " % genericsettings.inputsettings
+ 'argument for input flag "--settings".'
)
raise Usage(txt)
if 11 < 3:
from bbob_pproc import config # input settings
config.config(False)
import imp
# import testbedsettings as testbedsettings # input settings
try:
fp, pathname, description = imp.find_module("testbedsettings")
testbedsettings = imp.load_module("testbedsettings", fp, pathname, description)
finally:
fp.close()
if not genericsettings.verbose:
warnings.simplefilter("module")
# warnings.simplefilter('ignore')
# get directory name if outputdir is a archive file
algfolder = findfiles.get_output_directory_subfolder(args[0])
outputdir = os.path.join(outputdir, algfolder)
print ("Post-processing (1): will generate output " + "data in folder %s" % outputdir)
print " this might take several minutes."
filelist = list()
for i in args:
i = i.strip()
if os.path.isdir(i):
filelist.extend(findfiles.main(i, genericsettings.verbose))
elif os.path.isfile(i):
filelist.append(i)
else:
txt = "Input file or folder %s could not be found." % i
print txt
raise Usage(txt)
dsList = DataSetList(filelist, genericsettings.verbose)
if not dsList:
raise Usage("Nothing to do: post-processing stopped.")
if genericsettings.isNoisy and not genericsettings.isNoiseFree:
dsList = dsList.dictByNoise().get("nzall", DataSetList())
if genericsettings.isNoiseFree and not genericsettings.isNoisy:
dsList = dsList.dictByNoise().get("noiselessall", DataSetList())
# compute maxfuneval values
dict_max_fun_evals = {}
for ds in dsList:
dict_max_fun_evals[ds.dim] = np.max((dict_max_fun_evals.setdefault(ds.dim, 0), float(np.max(ds.maxevals))))
from bbob_pproc import config
config.target_values(genericsettings.isExpensive, dict_max_fun_evals)
config.config(dsList.isBiobjective())
if genericsettings.verbose:
for i in dsList:
if dict((j, i.instancenumbers.count(j)) for j in set(i.instancenumbers)) != inset.instancesOfInterest:
warnings.warn(
"The data of %s do not list " % (i) + "the correct instances " + "of function F%d." % (i.funcId)
)
dictAlg = dsList.dictByAlg()
if len(dictAlg) > 1:
warnings.warn("Data with multiple algId %s " % str(dictAlg.keys()) + "will be processed together.")
# TODO: in this case, all is well as long as for a given problem
# (given dimension and function) there is a single instance of
# DataSet associated. If there are more than one, the first one only
# will be considered... which is probably not what one would expect.
# TODO: put some errors where this case would be a problem.
# raise Usage?
if genericsettings.isFig or genericsettings.isTab or genericsettings.isRLDistr or genericsettings.isLogLoss:
if not os.path.exists(outputdir):
os.makedirs(outputdir)
if genericsettings.verbose:
print "Folder %s was created." % (outputdir)
if genericsettings.isPickled:
dsList.pickle(verbose=genericsettings.verbose)
if genericsettings.isConv:
ppconverrorbars.main(dictAlg, outputdir, genericsettings.verbose)
if genericsettings.isFig:
print "Scaling figures...",
sys.stdout.flush()
# ERT/dim vs dim.
plt.rc("axes", **inset.rcaxeslarger)
plt.rc("xtick", **inset.rcticklarger)
plt.rc("ytick", **inset.rcticklarger)
plt.rc("font", **inset.rcfontlarger)
plt.rc("legend", **inset.rclegendlarger)
plt.rc("pdf", fonttype=42)
ppfigdim.main(dsList, ppfigdim.values_of_interest, outputdir, genericsettings.verbose)
plt.rcdefaults()
print_done()
plt.rc("axes", **inset.rcaxes)
plt.rc("xtick", **inset.rctick)
plt.rc("ytick", **inset.rctick)
plt.rc("font", **inset.rcfont)
plt.rc("legend", **inset.rclegend)
plt.rc("pdf", fonttype=42)
if genericsettings.isTab:
print "TeX tables...",
sys.stdout.flush()
dictNoise = dsList.dictByNoise()
for noise, sliceNoise in dictNoise.iteritems():
pptable.main(sliceNoise, inset.tabDimsOfInterest, outputdir, noise, genericsettings.verbose)
print_done()
if genericsettings.isRLDistr:
print "ECDF graphs...",
sys.stdout.flush()
dictNoise = dsList.dictByNoise()
if len(dictNoise) > 1:
warnings.warn(
"Data for functions from both the noisy and "
"non-noisy testbeds have been found. Their "
'results will be mixed in the "all functions" '
"ECDF figures."
)
dictDim = dsList.dictByDim()
for dim in inset.rldDimsOfInterest:
try:
sliceDim = dictDim[dim]
except KeyError:
continue
pprldistr.main(sliceDim, True, outputdir, "all", genericsettings.verbose)
dictNoise = sliceDim.dictByNoise()
for noise, sliceNoise in dictNoise.iteritems():
pprldistr.main(sliceNoise, True, outputdir, "%s" % noise, genericsettings.verbose)
dictFG = sliceDim.dictByFuncGroup()
for fGroup, sliceFuncGroup in dictFG.items():
pprldistr.main(sliceFuncGroup, True, outputdir, "%s" % fGroup, genericsettings.verbose)
pprldistr.fmax = None # Resetting the max final value
pprldistr.evalfmax = None # Resetting the max #fevalsfactor
if (
genericsettings.isRldOnSingleFcts
): # copy-paste from above, here for each function instead of function groups
# ECDFs for each function
pprldmany.all_single_functions(dictAlg, None, outputdir, genericsettings.verbose)
print_done()
if genericsettings.isLogLoss:
print "ERT loss ratio figures and tables...",
sys.stdout.flush()
for ng, sliceNoise in dsList.dictByNoise().iteritems():
if ng == "noiselessall":
testbed = "noiseless"
elif ng == "nzall":
testbed = "noisy"
txt = "Please input crafting effort value " + "for %s testbed:\n CrE = " % testbed
CrE = genericsettings.inputCrE
while CrE is None:
try:
CrE = float(raw_input(txt))
except (SyntaxError, NameError, ValueError):
print "Float value required."
dictDim = sliceNoise.dictByDim()
for d in inset.rldDimsOfInterest:
try:
sliceDim = dictDim[d]
except KeyError:
continue
info = "%s" % ng
pplogloss.main(sliceDim, CrE, True, outputdir, info, verbose=genericsettings.verbose)
pplogloss.generateTable(sliceDim, CrE, outputdir, info, verbose=genericsettings.verbose)
for fGroup, sliceFuncGroup in sliceDim.dictByFuncGroup().iteritems():
info = "%s" % fGroup
pplogloss.main(sliceFuncGroup, CrE, True, outputdir, info, verbose=genericsettings.verbose)
pplogloss.evalfmax = None # Resetting the max #fevalsfactor
print_done()
latex_commands_file = os.path.join(outputdir.split(os.sep)[0], "bbob_pproc_commands.tex")
html_file = os.path.join(outputdir, genericsettings.single_algorithm_file_name + ".html")
prepend_to_file(
latex_commands_file, ["\\providecommand{\\bbobloglosstablecaption}[1]{", pplogloss.table_caption, "}"]
)
prepend_to_file(
latex_commands_file, ["\\providecommand{\\bbobloglossfigurecaption}[1]{", pplogloss.figure_caption, "}"]
)
prepend_to_file(
latex_commands_file,
[
"\\providecommand{\\bbobpprldistrlegend}[1]{",
pprldistr.caption_single(
np.max([val / dim for dim, val in dict_max_fun_evals.iteritems()])
), # depends on the config setting, should depend on maxfevals
"}",
],
)
replace_in_file(
html_file,
r"TOBEREPLACED",
"D, ".join([str(i) for i in pprldistr.single_runlength_factors[:6]]) + "D,…",
)
prepend_to_file(
latex_commands_file, ["\\providecommand{\\bbobppfigdimlegend}[1]{", ppfigdim.scaling_figure_caption(), "}"]
)
prepend_to_file(latex_commands_file, ["\\providecommand{\\bbobpptablecaption}[1]{", pptable.table_caption, "}"])
prepend_to_file(latex_commands_file, ["\\providecommand{\\algfolder}{" + algfolder + "/}"])
prepend_to_file(
latex_commands_file,
[
"\\providecommand{\\algname}{"
+ (str_to_latex(strip_pathname1(args[0])) if len(args) == 1 else str_to_latex(dsList[0].algId))
+ "{}}"
],
)
if genericsettings.isFig or genericsettings.isTab or genericsettings.isRLDistr or genericsettings.isLogLoss:
print "Output data written to folder %s" % outputdir
plt.rcdefaults()
def main(argv=None):
"""Generates some outputs from BBOB experiment data sets of two algorithms.
Provided with some data, this routine outputs figure and TeX files in the
folder 'cmp2data' needed for the compilation of the latex document
templateBBOBcmparticle.tex. These output files will contain performance
tables, performance scaling figures, scatter plot figures and empirical
cumulative distribution figures. On subsequent executions, new files will
be added to the output directory, overwriting existing files in the
process.
Keyword arguments:
argv -- list of strings containing options and arguments. If not given,
sys.argv is accessed.
argv must list folders containing BBOB data files. Each of these folders
should correspond to the data of ONE algorithm.
Furthermore, argv can begin with, in any order, facultative option flags
listed below.
-h, --help
display this message
-v, --verbose
verbose mode, prints out operations. When not in verbose mode, no
output is to be expected, except for errors.
-o, --output-dir OUTPUTDIR
change the default output directory ('cmp2data') to OUTPUTDIR
--noise-free, --noisy
restrain the post-processing to part of the data set only. Actually
quicken the post-processing since it loads only part of the pickle
files.
--fig-only, --rld-only, --tab-only, --sca-only
these options can be used to output respectively the ERT graphs
figures, run length distribution figures or the comparison tables
scatter plot figures only. Any combination of these options results
in no output.
Exceptions raised:
Usage -- Gives back a usage message.
Examples:
* Calling the runcomp2.py interface from the command line:
$ python bbob_pproc/runcomp2.py -v Alg0-baseline Alg1-of-interest
will post-process the data from folders Alg0-baseline and Alg1-of-interest,
the former containing data for the reference algorithm (zero-th) and the
latter data for the algorithm of concern (first). The results will be
output in folder cmp2data. The -v option adds verbosity.
* From the python interactive shell (requires that the path to this
package is in python search path):
>>> from bbob_pproc import runcomp2
>>> runcomp2.main('-o outputfolder PSO DEPSO'.split())
This will execute the post-processing on the data found in folder
PSO and DEPSO. The -o option changes the output folder from the default
cmp2data to outputfolder.
"""
if argv is None:
argv = sys.argv[1:]
# The zero-th input argument which is the name of the calling script is
# disregarded.
try:
try:
opts, args = getopt.getopt(argv, "hvo:", [
"help", "output-dir", "noisy", "noise-free", "fig-only",
"rld-only", "tab-only", "sca-only", "verbose"
])
except getopt.error, msg:
raise Usage(msg)
if not (args):
usage()
sys.exit()
isfigure = True
isrldistr = True
istable = True
isscatter = True
isNoisy = False
isNoiseFree = False # Discern noisy and noisefree data?
verbose = False
outputdir = 'cmp2data'
#Process options
for o, a in opts:
if o in ("-v", "--verbose"):
verbose = True
elif o in ("-h", "--help"):
usage()
sys.exit()
elif o in ("-o", "--output-dir"):
outputdir = a
elif o == "--fig-only":
isrldistr = False
istable = False
isscatter = False
elif o == "--rld-only":
isfigure = False
istable = False
isscatter = False
elif o == "--tab-only":
isfigure = False
isrldistr = False
isscatter = False
elif o == "--sca-only":
isfigure = False
isrldistr = False
istable = False
elif o == "--noisy":
isNoisy = True
elif o == "--noise-free":
isNoiseFree = True
else:
assert False, "unhandled option"
if (not verbose):
warnings.simplefilter('ignore')
print("BBOB Post-processing: will generate comparison " +
"data in folder %s" % outputdir)
print " this might take several minutes."
dsList, sortedAlgs, dictAlg = processInputArgs(args, verbose=verbose)
if not dsList:
sys.exit()
for i in dictAlg:
if isNoisy and not isNoiseFree:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'nzall', DataSetList())
if isNoiseFree and not isNoisy:
dictAlg[i] = dictAlg[i].dictByNoise().get(
'noiselessall', DataSetList())
for i in dsList:
if not i.dim in (2, 3, 5, 10, 20):
continue
#### The following lines are BBOB 2009 checking.###################
# Deterministic algorithms
#if i.algId in ('Original DIRECT', ):
#tmpInstancesOfInterest = instancesOfInterestDet
#else:
#tmpInstancesOfInterest = instancesOfInterest
#if ((dict((j, i.itrials.count(j)) for j in set(i.itrials)) <
#tmpInstancesOfInterest) and
#(dict((j, i.itrials.count(j)) for j in set(i.itrials)) <
#instancesOfInterest2010)):
#warnings.warn('The data of %s do not list ' %(i) +
#'the correct instances ' +
#'of function F%d or the ' %(i.funcId) +
#'correct number of trials for each.')
###################################################################
if (dict((j, i.itrials.count(j))
for j in set(i.itrials)) < instancesOfInterest2010):
warnings.warn('The data of %s do not list ' % (i) +
'the correct instances ' + 'of function F%d.' %
(i.funcId))
if len(sortedAlgs) < 2:
raise Usage('Expect data from two different algorithms, could ' +
'only find one.')
elif len(sortedAlgs) > 2:
#raise Usage('Expect data from two different algorithms, found ' +
# 'more than two.')
warnings.warn('Data from folders: %s ' % (sortedAlgs) +
'were found, the first two will be processed.')
# Group by algorithm
dsList0 = dictAlg[sortedAlgs[0]]
if not dsList0:
raise Usage('Could not find data for algorithm %s.' %
(sortedAlgs[0]))
#set_trace()
dsList1 = dictAlg[sortedAlgs[1]]
if not dsList1:
raise Usage('Could not find data for algorithm %s.' %
(sortedAlgs[0]))
tmppath0, alg0name = os.path.split(sortedAlgs[0].rstrip(os.sep))
tmppath1, alg1name = os.path.split(sortedAlgs[1].rstrip(os.sep))
#Trick for having different algorithm names in the tables...
#Does not really work.
#while alg0name == alg1name:
# tmppath0, alg0name = os.path.split(tmppath0)
# tmppath1, alg1name = os.path.split(tmppath1)
#
# if not tmppath0 and not tmppath1:
# break
# else:
# if not tmppath0:
# tmppath0 = alg0name
# if not tmppath1:
# tmppath1 = alg1name
#assert alg0name != alg1name
# should not be a problem, these are only used in the tables.
for i in dsList0:
i.algId = alg0name
for i in dsList1:
i.algId = alg1name
#for i, entry in enumerate(sortedAlgs): #Nota: key is sortedAlgs
#print "Alg%d is: %s" % (i, entry)
if isfigure or isrldistr or istable:
if not os.path.exists(outputdir):
os.mkdir(outputdir)
if verbose:
print 'Folder %s was created.' % (outputdir)
dictFN0 = dsList0.dictByNoise()
dictFN1 = dsList1.dictByNoise()
k0 = set(dictFN0.keys())
k1 = set(dictFN1.keys())
symdiff = k1 ^ k0
if symdiff: # symmetric difference
tmpdict = {}
for i, noisegrp in enumerate(symdiff):
if noisegrp == 'nzall':
tmp = 'noisy'
elif noisegrp == 'noiselessall':
tmp = 'noiseless'
if dictFN0.has_key(noisegrp):
tmp2 = sortedAlgs[0]
elif dictFN1.has_key(noisegrp):
tmp2 = sortedAlgs[1]
tmpdict.setdefault(tmp2, []).append(tmp)
txt = []
for i, j in tmpdict.iteritems():
txt.append('Only input folder %s lists %s data.' %
(i, ' and '.join(j)))
raise Usage('Data Mismatch: \n ' + ' '.join(txt) +
'\nTry using --noise-free or --noisy flags.')
if isfigure:
ppfig2.main(dsList0, dsList1, 1e-8, outputdir, verbose)
print "log ERT1/ERT0 vs target function values done."
if isrldistr:
if len(dictFN0) > 1 or len(dictFN1) > 1:
warnings.warn('Data for functions from both the noisy and ' +
'non-noisy testbeds have been found. Their ' +
'results will be mixed in the "all functions" ' +
'ECDF figures.')
dictDim0 = dsList0.dictByDim()
dictDim1 = dsList1.dictByDim()
for dim in set(dictDim0.keys()) | set(dictDim1.keys()):
if dim in rldDimsOfInterest:
try:
pprldistr2.main2(dictDim0[dim], dictDim1[dim],
rldValsOfInterest, outputdir,
'dim%02dall' % dim, verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.' %
(dim))
continue
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) | set(dictFG1.keys()):
pprldistr2.main2(dictFG0[fGroup], dictFG1[fGroup],
rldValsOfInterest, outputdir,
'dim%02d%s' % (dim, fGroup), verbose)
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) | set(dictFN1.keys()):
pprldistr2.main2(dictFN0[fGroup], dictFN1[fGroup],
rldValsOfInterest, outputdir,
'dim%02d%s' % (dim, fGroup), verbose)
print "ECDF absolute target graphs done."
#for dim in set(dictDim0.keys()) | set(dictDim1.keys()):
#if dim in rldDimsOfInterest:
#try:
#pprldistr2.main(dictDim0[dim], dictDim1[dim], None,
#True, outputdir, 'dim%02dall' % dim,
#verbose)
#except KeyError:
#warnings.warn('Could not find some data in %d-D.'
#% (dim))
#continue
#dictFG0 = dictDim0[dim].dictByFuncGroup()
#dictFG1 = dictDim1[dim].dictByFuncGroup()
#for fGroup in set(dictFG0.keys()) | set(dictFG1.keys()):
#pprldistr2.main(dictFG0[fGroup], dictFG1[fGroup], None,
#True, outputdir,
#'dim%02d%s' % (dim, fGroup), verbose)
#dictFN0 = dictDim0[dim].dictByNoise()
#dictFN1 = dictDim1[dim].dictByNoise()
#for fGroup in set(dictFN0.keys()) | set(dictFN1.keys()):
#pprldistr2.main(dictFN0[fGroup], dictFN1[fGroup],
#None, True, outputdir,
#'dim%02d%s' % (dim, fGroup), verbose)
#print "ECDF relative target graphs done."
for dim in set(dictDim0.keys()) | set(dictDim1.keys()):
pprldistr.fmax = None #Resetting the max final value
pprldistr.evalfmax = None #Resetting the max #fevalsfactor
if dim in rldDimsOfInterest:
try:
pprldistr.comp(dictDim0[dim], dictDim1[dim],
rldValsOfInterest, True, outputdir,
'dim%02dall' % dim, verbose)
except KeyError:
warnings.warn('Could not find some data in %d-D.' %
(dim))
continue
dictFG0 = dictDim0[dim].dictByFuncGroup()
dictFG1 = dictDim1[dim].dictByFuncGroup()
for fGroup in set(dictFG0.keys()) | set(dictFG1.keys()):
pprldistr.comp(dictFG0[fGroup], dictFG1[fGroup],
rldValsOfInterest, True, outputdir,
'dim%02d%s' % (dim, fGroup), verbose)
dictFN0 = dictDim0[dim].dictByNoise()
dictFN1 = dictDim1[dim].dictByNoise()
for fGroup in set(dictFN0.keys()) | set(dictFN1.keys()):
pprldistr.comp(dictFN0[fGroup], dictFN1[fGroup],
rldValsOfInterest, True, outputdir,
'dim%02d%s' % (dim, fGroup), verbose)
print "ECDF dashed-solid graphs done."
if istable:
dictFN0 = dsList0.dictByNoise()
dictFN1 = dsList1.dictByNoise()
for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
pptable2.mainnew(dictFN0[fGroup], dictFN1[fGroup],
tabDimsOfInterest, outputdir, '%s' % (fGroup),
verbose)
#pptable2.main2(dsList0, dsList1, tabDimsOfInterest, outputdir,
# verbose=verbose)
if isscatter:
ppscatter.main(dsList0, dsList1, outputdir, verbose=verbose)
if isfigure or isrldistr or istable or isscatter:
print "Output data written to folder %s." % outputdir