cocopp.cococommands

module documentation

A collection of commands for using COCO from Jupyter or IPython.

The main data structures used in COCO are DataSet, which corresponds to data of one algorithm on one problem, and DataSetList, which is for collections of DataSet instances. Both classes are implemented in cocopp.pproc.

Examples:

Start by importing cocopp:

>>> import cocopp
>>> cocopp.genericsettings.verbose = False # ensure to make below doctests work
>>> def print_(*args, **kwargs): pass
>>> cocopp.archives.bbob._print = print_  # avoid download notification

Load a data set, assign to variable ds:

>>> path = cocopp.archives.bbob.get(4)
>>> print('ESC'); dsl = cocopp.load(path)  # a dataset list  # doctest:+ELLIPSIS
ESC...
>>> ds = dsl[0]

Get some information on a DataSetList instance:

>>> print(dsl)  # doctest:+ELLIPSIS
[DataSet(BIPOP-CMA-ES_hansen on f1 2-D), DataSet(BIPOP-CMA-ES...
>>> dsl.info()
144 data set(s)
Algorithm(s): BIPOP-CMA-ES_hansen
24 Functions with IDs 1-24
Dimension(s): 2, 3, 5, 10, 20, 40
Max evals: [1625595, 2349823, 3114271, 5884514, 12102699, 36849608]

Get some information on a DataSet instance:

>>> print(ds)
DataSet(BIPOP-CMA-ES_hansen on f1 2-D)
>>> ds.info()
Algorithm: BIPOP-CMA-ES_hansen
Function ID: 1
Dimension DIM = 2
Number of trials: 15
Final target Df: 1e-08
min / max number of evals per trial: 224 / 333
   evals/DIM:  best     15%     50%     85%     max |  ERT/DIM  nsucc
  ---Df---|-----------------------------------------|----------------
  1.0e+03 |       0       0       0       0       0 |      0.5  15
  1.0e+01 |       0       0       2       8      10 |      2.9  15
  1.0e-01 |       8      13      22      38      52 |     24.2  15
  1.0e-03 |      34      48      56      74      77 |     58.2  15
  1.0e-05 |      64      70      89     100     102 |     86.1  15
  1.0e-08 |     112     116     128     150     166 |    130.9  15

Class	`DataWithFewSuccesses`	This class is superseded by `cocoex.ExperimentRepeater` for its main usecase,
Function	`info`	Display more info on an instance of DatasetList.
Function	`load`	[currently broken when further used within `cocopp`, see `load2`] Create a `DataSetList` instance from a file or folder.
Function	`load2`	[WIP] return a `DataSetList` or a `dict` of `dict` of `DataSetLists`.
Function	`systeminfo`	Display information on the system.
Function	`true_number_of_trials`	return number of actually conducted trials for `DataSet` `ds`
Function	`_pickle`	Pickle a DataSetList.

def info(dsList): ¶

Display more info on an instance of DatasetList.

def load(filename): ¶

[currently broken when further used within cocopp, see load2] Create a DataSetList instance from a file or folder.

Input argument filename can be a single info file name, a single pickle filename or a folder name. In the latter case, the folder is browsed recursively for info or pickle files.

Details: due to newly implemented side effects when data are read in, the returned data set list may not work anymore when used with plotting functions of the cocopp module, see also load2.

def load2(args, keep=None, remove=None, flat_list=None): ¶

[WIP] return a DataSetList or a dict of dict of DataSetLists.

args is a string or a list of strings passed to cocopp.official_archives.all.get_extended to determine the desired data sets which can also come from a local folder or a zip-file.

keep and remove are callables taking a cocopp.pproc.DataSet class instance as argument and return a bool according to which data sets are filtered (usually based on the funcId attribute, see below example). In case, remove takes precedent. When keep is None all elements are kept unless removed.

return a DataSetList if either flat_list or flat_list is None and args describes a single algorithm/experiment, return otherwise a dict of dict with dimension and pathname as keys, respectively.

Examples:

# Get a single `DataSet`, here for F1 in 20-D
>> ds = cocopp.load2('nelder!').by('dim')[20].by('funcId')[1][0]

# get 31 data sets, takes ~3 minutes at first ever loading
>> ddsl31 = load('bbob/2009/*')
>> assert sorted(ddsl31) == [2, 3, 5, 10, 20, 40], ddsl
>> assert len(ddsl31[3]) == 31, ddsl

>>> import cocopp
>>> def pprldmany(dsl):
...     print('OK')  # circumvent output checking
...     with cocopp.toolsdivers.InfolderGoneWithTheWind():
...         cocopp.compall.pprldmany.main(dsl)  # writes pprldmany_default.*
...
>>> _, ddsl = (print('OK'),  # circumvent output checking
...     cocopp.load2('bbob/2009/B*',
...                  remove=lambda ds: ds.funcId in [1, 5]))  # doctest:+ELLIPSIS
OK...
>>> assert sorted(ddsl) == [2, 3, 5, 10, 20, 40], sorted(ddsl)
>>> assert all([len(ddsl[i]) == 3 for i in ddsl]), ddsl  # 3 algorithms
>>> assert len(list(ddsl[3].values())[0]) == 22, list(ddsl[3].values())[0]
>>> pprldmany(ddsl[2])  # doctest:+ELLIPSIS
OK...

def systeminfo(): ¶

Display information on the system.

def true_number_of_trials(ds): ¶

return number of actually conducted trials for DataSet ds

def _pickle(dsList): ¶

Pickle a DataSetList.