2.2.2

New Features

• more prominent and improved documentation of how to use the archived data
• BBOB-2018 data added to archived data
• improved ECDF graphs (increased readability of labels & use of space, shortened x-axis, lighter cross markers)
• cocopp.load can load archived and mixed data
• cocopp.main opens webbrowser automatically when post-processing is finished
• added bbob-new2 data format description (mainly for yet unsupported constrained problems)

Bug Fixes

• bug-fix when calling python -m cocopp.test
• fixed distribution of initial solution for restarts in example_experiment.py
• displayed number of instances not depending on reference algorithm anymore (#1792)
• bug-fix in example_experiment2.py when calling CMA-ES

Others

• replaced cmocka with minunit
• file-flush in logger now at the right place and for all files
• changes in (yet unsupported) bbob-constrained test suite
• more parameters in runlength-based setting accessible through genericsettings.py (#1284)
• dataformatsetting removed from testbedsetting
• regression test files are automatically downloaded and not in the repository anymore

2.2.1

Bug fixes

• Labels in ECDF plots not connected with graph (#1724)
• UserWarning in postprocessing due to wrong setting of instancesOfInterest for bbob suite

2.2

Official release to do experiments for the BBOB-2018 workshop.

New features

• automated download of algorithm data sets via cocopp.data_archive (#1533)
• updated convergence plots (#1113)
• LaTeX tables are reorganized to have better control within LaTeX (#1376)
• Noisy LaTeX template transferred from svn repository to github and updated (#1171)
• improved Problem.info string in python wrapper
• cocopp modules offers a set_seed
• postprocessing cocopp module now python 3 compatible
• all figures (and their inclusion to the LaTeX templates) have been updated and beautified in order to comply with the newest matplotlib version
• zip files are supported as input files (#35)
• improved handling of input parameters to suite (#863)
• postprocessing results for more than one algorithm are now also written to a subfolder

Interface Changes

• The reference worst f-values-of-interest are exposed to the (multiobjective) solver (#1120)
• The name cocoex.Problem.list_of_observers has changed to cocoex.Problem.observers
• Sole entry points to the postprocessing are main and rungeneric.main. Got rid of rungeneric2.py and removed pptable2.py
• cocopp.main now returns a DataSetList
• Algorithms can now be displayed in the background (#1370)
• COCO problems have now the is_observed property (#1388)
• COCO observer allows to query the output folder where it wrote the results (#1425)
• Extracted data is now written to .extracted_... folders instead of _extracted_... to allow using glob.glob(*) as inputs even when data is already extracted.
• The former --omit-single functionality became now the default. To use old functionality, use the new --include-single switch.
• Simplified example_experiment_for_beginners.py available
• New non-anytime example experiment in python to benchmark budget-dependent algorithms
• Single-objective experiments now log also the number of constraint evaluations

Important bug fixes

• error with Suite.free (#315)
• warnings due to unclosed files
• corrected wrong statements in some captions with respect to Bonferroni correction
• corrected best parameter update in some of the transformations (#1695)
• corrected erroneous markers in one algorithm ECDF (#1670)

Misc

• removed several compiler warnings (#1350)
• reorganization of postprocessing's output files
• cleanup of postprocessing: removing old, unused files
• made max path long enough to actually hold a path of maximal length
• continuous integration testing expaned to CircleCI and AppVeyor
• minor cleanup of do.py
• small format adjustment in tables (#1689)
• updated doctests in postprocessing to fit a change in numpy (#1703)

2.1.1

Fixes a few important bugs, in particular with respect to incompatibility with the latest matplotlib versions.

2.1

Release for preparing the final submissions to BBOB2017.

New features

• Tables generated for all dimensions.
• Added a few more line styles to better distinguish the 31 algorithms of BBOB-2009.

Important bug fixes

• Fixed the instance order in the best algorithm info file.
• Reference to table(s) in biobjective LaTeX template corrected.
• Corrected layout of single-algorithm tables in LaTeX.
• Correct reference algorithm for bbob-biobj suite from release 2.0 restored (release 2.0.1 had for some reason a wrong reference algorithm).
• Bug while constructing the reference hypervolume hash when the original data is in too many separate files.

2.0.1

New features

• LaTeX templates now all in new ACM style, requested for submitting final papers to BBOB-2017 (#1272)
• Tables in single-algorithm LaTeX template are now split into per-function tables that are easier to move and adjust for new (larger) test suites
• improved human-readable information available within the best/reference algorithms

Important bug fixes

• best/reference algorithms were not usable in version 2.0 (see for example #1327)
• wrong alignment of success probability in tables (#1309)

2.0

Main release after the BBOB-2016 workshop, including updated hypervolume reference values from the 15 bi-objective submissions to the workshop.

New features

• The biggest change affects the postprocessing, the call to which changed its name: now called via python -m cocopp [options] ALGO1 [OPTIONAL ALGOs]
• Updated hypervolume reference values for the bbob-biobj test suite, based in particular on all non-dominated solutions found by the submitted BBOB-2016 workshop algorithms DEMO, GA-MULTIOBJ-NSGA-II, HMO-CMA-ES, MAT-DIRECT, MAT-SMS, MO-DIRECT-HV-Rank, MO-DIRECT-ND, MO-DIRECT-Rank, RM-MEDA, RS-4, RS-5, RS-100, SMS-EMOA-DE, SMS-EMOA-PM, and UP-CMA-ES.
• The data of all biobjective BBOB-2016 algorithm data sets are available online at the same time at http://coco.gforge.inria.fr/data-archive/bbob-biobj/2016/
• A Best 2016 algorithm is now available in the figures for the bbob-biobj test suite as the artificial best algorithm from all 15 submitted BBOB-2016 algorithms on that suite.
• Any algorithm data set can be used as reference algorithm in the postprocessing by specifying the filename in the corresponding testbed class in testbedsettings.py. At the same time, pickled best algorithms are not supported anymore.
• New extended biobjective suite bbob-biobj-ext, extending the bbob-biobj suite from 55 to 92 functions.
• New instances as default, in particular now 15 instances to be run for the biobjective test suites as well.
• Improved HTML output and navigation
• Updated documentation
• Updated and cleaned grayscale settings for producing grayscale figures
• More information is shown in the figures and tables, in particular the Coco version number and a hash of the used hypervolume reference values in the biobjective case in order to be able to compare plots over different papers (plots are comparable only if the hash is the same).
• New regression tests for all provided test suites
• Reduced the number of ppscatter_target_values to 21 for better readability

Important bug fixes

• Bug in python wrapper reported by S. Wessing (075d2fb)
• solved issues #1206, #1232, #1236, #1265 (amongst others)

1.2.1

New features

• Generalized Sharp Ridge Function for larger dimension (#1093)
• Source code of documentation moved to separate coco-doc repository (#1129)
• arXiv papers linked in LaTeX templates (#1003)
• Global setting of verbosity in postprocessing (#1042)
• New tests for archive updates (#843)
• First integration test for bbob-noisy data sets
• Improved Python example experiment with interface change

Important bug fixes

• HTML tables corrected for bi-objective many algorithm case (#1109)
• Bug in doctest in python wrapper of Suite.current_index fixed

1.2

Minor release with few new features and several bugfixes, including the hotfix for the important issue #1141.

New features

• Smallest and largest target values displayed in figures together with their number (#1122)
• Improved documentation on how to build a new test suite (#92)
• Improved test coverage of bbob test suite (#852)
• Shortened text on number of instances in figures for many algorithms (#1077 and #1115)

Important bug fixes

• Wrong placement of new coco_version in data files (#1141)
• Standard colors got overwritten within many algorithm comparison (#1134)
• Potential division by zero in about_equal_value (#1126)
• BibTeX file now included in postprocessing installation (#335)
• Momentarily reduced the warnings shown to the user due to a missing best_parameter update which, however, does not affect any result for now (#1125)
• Corrected some warnings

1.1.4

New features

• Now the archives work internally with discretized normalized values.
  • Note here that the usage of not discretized values in the past might have caused algorithms to appear slightly better or worse than they actually are in the BBOB-2016 submissions.
• Added version number to the experiments and preprocessing output.
• Next dimension links in html output use just the existing dimensions (#1067).
• Added verbose option to do.py in order to print more output to the console.

Important bug fixes

• The distance to the ROI was not correctly computed for some solutions.
• Running postprocessing with many algorithms is now possible as well in grayscale setting (reported by @melovv in #1077)