Example #1
0
def main(argv=None):
    r"""Main routine for post-processing the data of multiple algorithms.

    Provided with some data, this routine outputs figure and TeX files
    in a folder needed for the compilation of latex document
    :file:`template3XXX.tex` or :file:`noisytemplate3XXX.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
    provided, sys.argv is accessed.

    *argv* must list folders containing BBOB data files.
    The name of these folders will be used in the output figures and
    tables to designate the algorithms. Therefore you should name the
    folders with differentiating names.

    Furthermore, argv can begin with facultative option flags listed
    below.

        -h, --help
            displays this message.
        -v, --verbose
            verbose mode, prints out operations, warnings.
        -o OUTPUTDIR, --output-dir=OUTPUTDIR
            changes the default output directory (:file:`ppdatamany`) to
            :file:`OUTPUTDIR`.
        --noise-free, --noisy
            processes only part of the data.
        --settings=SETTINGS
            changes the style of the output figures and tables. At the
            moment only 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, --rld-only, --fig-only
            these options can be used to output respectively the
            comparison TeX tables, the run lengths distributions or the
            figures of ERT/dim vs dim only. A combination of any two or
            more of these options results in no output.
        --conv
            if this option is chosen, additionally convergence
            plots for each function and algorithm are generated.
        --perf-only
            generate only performance plots
        --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. 
        -

    Exceptions raised:

    *Usage* -- Gives back a usage message.

    Examples:

    * Calling the rungenericmany.py interface from the command line::

        $ python bbob_pproc/rungenericmany.py -v AMALGAM BFGS BIPOP-CMA-ES


    * 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.rungenericmany -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.rungenericmany.main('-o outputfolder folder1 folder2'.split())

      This will execute the post-processing on the data found in
      :file:`folder1` and :file:`folder2`.
      The ``-o`` option changes the output folder from the default to
      :file:`outputfolder`.

    * Generate post-processing data for some algorithms with figures in
      shades of gray::

        $ python rungenericmany.py --settings grayscale NEWUOA NELDER LSSTEP

    """

    if argv is None:
        argv = sys.argv[1:]

    try:
        try:
            opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
        except getopt.error, msg:
            raise Usage(msg)

        if not (args):
            usage()
            sys.exit()

        verbose = False
        outputdir = 'ppdata'
        isNoisy = False
        isNoiseFree = False

        isPer = True
        isTab = True
        isFig = True
        inputsettings = "color"
        isConv = False
        isRLbased = None  # allows automatic choice
        isExpensive = None 
        isRldOnSingleFcts = 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 == "--noisy":
                isNoisy = True
            elif o == "--noise-free":
                isNoiseFree = True
            #The next 3 are for testing purpose
            elif o == "--tab-only":
                isPer = False
                isFig = False
            elif o == "--rld-single-fcts":
                isRldOnSingleFcts = True
            elif o == "--rld-only":
                isTab = False
                isFig = False
            elif o == "--fig-only":
                isPer = False
                isTab = False
            elif o == "--perf-only":
                isTab = False
                isFig = False
            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
        # TODO: conditional imports are NOT the way to go here
        if inputsettings == "color":
            from bbob_pproc import config, genericsettings as inset # input settings
            config.config()
        elif inputsettings == "grayscale":
            # this settings strategy (by proving different settings files) is problematic, 
            # because it means copy-paste of the settings
            # file and future changes have a great chance to make the pasted files incompatible
            # as has most likely happened with grayscalesettings:
            from bbob_pproc import config, grayscalesettings as inset # input settings
            # better would be just adjust the previous settings, as config is doing it, 
            # so a config_grayscalesettings.py module seems the better approach to go 
        elif inputsettings == "black-white":
            from bbob_pproc import config, 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.filterwarnings('module', '.*', Warning, '.*')  # same warning just once
            warnings.simplefilter('ignore')  # that is bad, but otherwise to many warnings appear 

        if isRLbased is not None:
            genericsettings.runlength_based_targets = isRLbased
        config.target_values(isExpensive)
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 aRTs graphs figures, run length
            distribution figures, aRT 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.
        --no-rld-single-fcts
            do not generate runlength distribution figures for each
            single function.
        --expensive
            runlength-based f-target values and fixed display limits,
            useful with comparatively small budgets.
        --no-svg
            do not generate 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 == "--no-rld-single-fcts":
                genericsettings.isRldOnSingleFcts = False
            elif o == "--runlength-based":
                genericsettings.runlength_based_targets = True
            elif o == "--expensive":
                genericsettings.isExpensive = True  # comprises runlength-based
            elif o == "--no-svg":
                genericsettings.generate_svg_files = False
            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 . import genericsettings as inset  # input settings
        elif genericsettings.inputsettings == "grayscale":
            from . import grayscalesettings as inset  # input settings
        elif genericsettings.inputsettings == "black-white":
            from . 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()
            import imp
            # import testbedsettings as testbedsettings # input settings
            try:
                fp, pathname, description = imp.find_module("testbedsettings")
                testbedsettings1 = 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("\nPost-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. For more information check the messages above.")

        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 . import config
        config.target_values(genericsettings.isExpensive)
        config.config(dsList[0].testbed_name())

        if (genericsettings.verbose):
            for i in dsList:                
                # check whether current set of instances correspond to correct
                # setting of a BBOB workshop and issue a warning otherwise:            
                curr_instances = (dict((j, i.instancenumbers.count(j)) for j in set(i.instancenumbers)))
                correct = False
                for instance_set_of_interest in inset.instancesOfInterest:
                    if curr_instances == instance_set_of_interest:
                        correct = True
                if not correct:
                    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.

        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:
            print("Generating convergence plots...")
            ppconverrorbars.main(dictAlg, 
                                 dsList.isBiobjective(),
                                 outputdir, 
                                 genericsettings.verbose,
                                 genericsettings.single_algorithm_file_name)
            print_done()

        values_of_interest = testbedsettings.current_testbed.ppfigdim_target_values
        if genericsettings.isFig:
            print("Scaling figures...")
            # aRT/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, 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("Generating LaTeX tables...")
            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...")
            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

                dictNoise = sliceDim.dictByNoise()

                # If there is only one noise type then we don't need the all graphs.
                if len(dictNoise) > 1:
                    pprldistr.main(sliceDim, True,
                                   outputdir, 'all', genericsettings.verbose)
                
                    
                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
            print_done()

            if genericsettings.isRldOnSingleFcts: # copy-paste from above, here for each function instead of function groups
                # ECDFs for each function
                print("ECDF graphs per function...")
                pprldmany.all_single_functions(dictAlg, 
                                               dsList.isBiobjective(),
                                               True,
                                               None,
                                               outputdir,
                                               genericsettings.verbose,
                                               genericsettings.single_algorithm_file_name)
                print_done()
            
        if genericsettings.isLogLoss:
            print("aRT loss ratio figures and tables...")
            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()

        dictFunc = dsList.dictByFunc()
        ppfig.save_single_functions_html(os.path.join(outputdir, genericsettings.single_algorithm_file_name),
                                    dictFunc[dictFunc.keys()[0]][0].algId,
                                    htmlPage = ppfig.HtmlPage.ONE,
                                    isBiobjective = dsList.isBiobjective(),
                                    functionGroups = dsList.getFuncGroups())

        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(),  # depends on the config setting, should depend on maxfevals
                         '}'])
        html_file = os.path.join(outputdir, 'pprldistr.html')
        replace_in_file(html_file, r'TOBEREPLACED', 'D, '.join([str(i) for i in pprldistr.single_runlength_factors[:6]]) + 'D,&hellip;')
        prepend_to_file(latex_commands_file,
                        ['\\providecommand{\\bbobppfigdimlegend}[1]{',
                         ppfigdim.scaling_figure_caption(),
                         '}'])
        prepend_to_file(latex_commands_file,
                        ['\\providecommand{\\bbobpptablecaption}[1]{',
                         pptable.get_table_caption(),
                         '}'])
        prepend_to_file(latex_commands_file,
                        ['\\providecommand{\\bbobecdfcaptionsinglefcts}{',
                         ppfigs.get_ecdfs_single_fcts_caption(),
                         '}'])
        prepend_to_file(latex_commands_file,
                        ['\\providecommand{\\bbobecdfcaptionallgroups}{',
                         ppfigs.get_ecdfs_all_groups_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()
Example #3
0
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()
Example #4
0
            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
        config.target_values(isExpensive, dict_max_fun_evals)
        config.config()


        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))

        plt.rc("axes", **inset.rcaxes)
        plt.rc("xtick", **inset.rctick)
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`
        --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. 

    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, 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
        verbose = False
        outputdir = 'ppdata'
        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 ("-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 == "--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
            config.config()
        elif inputsettings == "grayscale": # probably very much obsolete
            from bbob_pproc import grayscalesettings as inset # input settings
        elif inputsettings == "black-white": # probably very much obsolete
            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:
                dictAlg[i] = dictAlg[i].dictAll().get('All', 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%s.' %(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))))
        if isRLbased is not None:
            genericsettings.runlength_based_targets = isRLbased
        config.target_values(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 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.dictAll()
        # dictFN1 = dsList1.dictAll()
        # 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, ppfig2_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], 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], dim,
                                        inset.rldValsOfInterest,
                                        outputdir, '%02dD_%s' % (dim, fGroup),
                                        verbose)

                    # ECDFs for all
                    dictFN0 = dictDim0[dim].dictAll()
                    dictFN1 = dictDim1[dim].dictAll()

                    for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
                        pprldistr2.main(dictFN1[fGroup], dictFN0[fGroup], dim,
                                        inset.rldValsOfInterest,
                                        outputdir, '%02dD_%s' % (dim, fGroup),
                                        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', 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, verbose)

                    # ECDFs for all
                    dictFN0 = dictDim0[dim].dictAll()
                    dictFN1 = dictDim1[dim].dictAll()
                    for fGroup in set(dictFN0.keys()) & set(dictFN1.keys()):
                        pprldistr.comp(dictFN1[fGroup], dictFN0[fGroup],
                                       inset.rldValsOfInterest, 
                                       True, outputdir,
                                       '%s' % fGroup, verbose)

            print "ECDF runlength graphs done."

        if isConv:
            ppconverrorbars.main(dictAlg,outputdir,verbose)

##TODO pptable
#        if istable:
#             dictNG0 = dsList0.dictAll()
#             dictNG1 = dsList1.dictAll()
#
#             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)
#
#             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,
#                              '}'
#                             ])
#             print "Tables done."

        if isscatter:
            if genericsettings.runlength_based_targets:
                ppscatter.targets = ppscatter.runlength_based_targets
            ppscatter.main(dsList1, dsList0, outputdir, verbose=verbose)
            prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'), 
                            ['\\providecommand{\\bbobppscatterlegend}[1]{', 
                             ppscatter.figure_caption(), 
                             '}'
                            ])
            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)
            if genericsettings.runlength_based_targets:
                ftarget = RunlengthBasedTargetValues([target_runlength])  # TODO: make this more variable but also consistent
            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()
Example #6
0
            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())



        # 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))))
            
        # set target values
        from bbob_pproc import config
        config.target_values(genericsettings.isExpensive, dict_max_fun_evals)
        config.config(dsList[0].isBiobjective())


        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))

        plt.rc("axes", **inset.rcaxes)
        plt.rc("xtick", **inset.rctick)
def main(argv=None):
    r"""Main routine for post-processing the data of multiple algorithms.

    Provided with some data, this routine outputs figure and TeX files
    in a folder needed for the compilation of latex document
    :file:`template3XXX.tex` or :file:`noisytemplate3XXX.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
    provided, sys.argv is accessed.

    *argv* must list folders containing BBOB data files.
    The name of these folders will be used in the output figures and
    tables to designate the algorithms. Therefore you should name the
    folders with differentiating names.

    Furthermore, argv can begin with facultative option flags listed
    below.

        -h, --help
            displays this message.
        -v, --verbose
            verbose mode, prints out operations, warnings.
        -o OUTPUTDIR, --output-dir=OUTPUTDIR
            changes the default output directory (:file:`ppdatamany`) to
            :file:`OUTPUTDIR`.
        --noise-free, --noisy
            processes only part of the data.
        --settings=SETTINGS
            changes the style of the output figures and tables. At the
            moment only 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, --rld-only, --fig-only
            these options can be used to output respectively the
            comparison TeX tables, the run lengths distributions or the
            figures of ERT/dim vs dim only. A combination of any two or
            more of these options results in no output.
        --conv
            if this option is chosen, additionally convergence
            plots for each function and algorithm are generated.
        --perf-only
            generate only performance plots
        --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. 
        -

    Exceptions raised:

    *Usage* -- Gives back a usage message.

    Examples:

    * Calling the rungenericmany.py interface from the command line::

        $ python bbob_pproc/rungenericmany.py -v AMALGAM BFGS BIPOP-CMA-ES


    * 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.rungenericmany -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.rungenericmany.main('-o outputfolder folder1 folder2'.split())

      This will execute the post-processing on the data found in
      :file:`folder1` and :file:`folder2`.
      The ``-o`` option changes the output folder from the default to
      :file:`outputfolder`.

    * Generate post-processing data for some algorithms with figures in
      shades of gray::

        $ python rungenericmany.py --settings grayscale NEWUOA NELDER LSSTEP

    """

    if argv is None:
        argv = sys.argv[1:]

    try:
        try:
            opts, args = getopt.getopt(argv, shortoptlist, longoptlist)
        except getopt.error, msg:
            raise Usage(msg)

        if not (args):
            usage()
            sys.exit()

        verbose = False
        outputdir = 'ppdata'
        isNoisy = False
        isNoiseFree = False

        isPer = True
        isTab = True
        isFig = True
        inputsettings = "color"
        isConv = False
        isRLbased = None  # allows automatic choice
        isExpensive = None 
        isRldOnSingleFcts = 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 == "--noisy":
                isNoisy = True
            elif o == "--noise-free":
                isNoiseFree = True
            #The next 3 are for testing purpose
            elif o == "--tab-only":
                isPer = False
                isFig = False
            elif o == "--rld-single-fcts":
                isRldOnSingleFcts = True
            elif o == "--rld-only":
                isTab = False
                isFig = False
            elif o == "--fig-only":
                isPer = False
                isTab = False
            elif o == "--perf-only":
                isTab = False
                isFig = False
            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
        # TODO: conditional imports are NOT the way to go here
        if inputsettings == "color":
            from bbob_pproc import config, genericsettings as inset # input settings
            config.config()
        elif inputsettings == "grayscale":
            # this settings strategy (by proving different settings files) is problematic, 
            # because it means copy-paste of the settings
            # file and future changes have a great chance to make the pasted files incompatible
            # as has most likely happened with grayscalesettings:
            from bbob_pproc import config, grayscalesettings as inset # input settings
            # better would be just adjust the previous settings, as config is doing it, 
            # so a config_grayscalesettings.py module seems the better approach to go 
        elif inputsettings == "black-white":
            from bbob_pproc import config, 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.filterwarnings('module', '.*', Warning, '.*')  # same warning just once
            warnings.simplefilter('ignore')  # that is bad, but otherwise to many warnings appear 

        if isRLbased is not None:
            genericsettings.runlength_based_targets = isRLbased
        config.target_values(isExpensive)
            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
        config.target_values(isExpensive, dict_max_fun_evals)
        config.config()


        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))

        plt.rc("axes", **inset.rcaxes)
        plt.rc("xtick", **inset.rctick)
                    'nzall', DataSetList())
            if genericsettings.isNoiseFree and not genericsettings.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))))

        # set target values
        from bbob_pproc import config
        config.target_values(genericsettings.isExpensive, dict_max_fun_evals)
        config.config()

        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))

        plt.rc("axes", **inset.rcaxes)
        plt.rc("xtick", **inset.rctick)
Example #10
0
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()
Example #11
0
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.
        --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. 

    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, 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
        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 ("-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
            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
            config.config()
        elif inputsettings == "grayscale":  # probably very much obsolete
            from bbob_pproc import grayscalesettings as inset  # input settings
        elif inputsettings == "black-white":  # probably very much obsolete
            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

        # 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))))
        if isRLbased is not None:
            genericsettings.runlength_based_targets = isRLbased
        config.target_values(
            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 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, ppfig2_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], 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], dim,
                                        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], dim,
                                        inset.rldValsOfInterest, outputdir,
                                        '%02dD_%s' % (dim, fGroup), 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',
                            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, 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, 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)

            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, '}'
                    ])
            print "Tables done."

        if isscatter:
            if genericsettings.runlength_based_targets:
                ppscatter.targets = ppscatter.runlength_based_targets
            ppscatter.main(dsList1, dsList0, outputdir, verbose=verbose)
            prepend_to_file(os.path.join(outputdir, 'bbob_pproc_commands.tex'),
                            [
                                '\\providecommand{\\bbobppscatterlegend}[1]{',
                                ppscatter.figure_caption(), '}'
                            ])
            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)
            if genericsettings.runlength_based_targets:
                ftarget = RunlengthBasedTargetValues([
                    target_runlength
                ])  # TODO: make this more variable but also consistent
            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()