|Pypi Version| |Build Status| |Coverage Status| |Codacy Status|
Whatami is an attempt to abstract configurability and experiment identifiability in a convenient way. It does so by leveraging object introspection and defining a simple, flexible and consistent API that python objects can adhere to, even if they were not originally designed to do it.
It works this way:
Objects provide their own ids based on parameters=value dictionaries. They do so by returning an instance of the What class from a method called "what()".
Optionally, this package provides a whatable decorator to provide automatic creation of What objects from objects dictionaries, slots and properties. All attributes will be considered part of the configuration, except for those whose names start or end by '_'.
They aim to look like they would be generated if repr in python was always implemented taking into account recursion and exposing only result-changing parameters. They pretty much look like python function calls with nested parameters expanded to look like python calls themselves.
Represent your computations as standardized ID strings.
Pluggable architecture.
ID strings can be parsed and manipulated.
"whatamise" your library. Included support for:
scikit-learn_pyopy_Convenience functions for data tidying.
Whatami is simple but powerful. This example just shows its surface and better docs are coming, but in the meantime, just check the docstrings and unit tests.
.. code:: python
# Objects of this class provide a configuration (`What` object)
class DuckedConfigurable(object):
def __init__(self, quantity, name, company=None, verbose=True):
self.quantity = quantity
self.name = name
self.company = company
self.verbose = verbose
def what(self):
return What('ducked', {'quantity': self.quantity, 'name': self.name, 'company': self.company})
duckedc = DuckedConfigurable(33, 'salty-lollypops', verbose=False)
# The configuration id string sorts by key alphanumeric order, helping id consistency
print duckedc.what().id()
# ducked(company=None,name='salty-lollypops',quantity=33)
# Using the whatable decorator makes objects gain a what() method;
# in this case, what() is infered automatically
@whatable
class Company(object):
def __init__(self, name, city, verbose=True):
super(Company, self).__init__()
self.name = name
self.city = city
self._verbose = verbose # not part of config
self.social_reason_ = '%s S.A., %s' % (name, city) # not part of config
cc = Company(name='Chupa Chups', city='Barcelona')
print(cc.what().id())
# Company(city='Barcelona',name='Chupa Chups')
# Ultimately, we can nest whatables...
duckedc = DuckedConfigurable(33, 'salty-lollypops', company=cc, verbose=False)
print duckedc.what().id()
# ducked(company=Company(city='Barcelona',name='Chupa Chups'),name='salty-lollypops',quantity=33)
# We can also decorate functions and partials - use with caution
@whatable
def buy(company, price=2**32, currency='euro'):
return '%s is now mine for %g %s' % (company.name, price, currency)
print buy.what().id()
# buy(currency='euro',price=4294967296)Since release 4.0.0 whatami uses semantic versioning_, where a major version bump
happens also if the default id strings can be generated differently, even if no API
actually changes.
.. |Build Status| image:: https://travis-ci.org/sdvillal/whatami.svg?branch=master :target: https://travis-ci.org/sdvillal/whatami .. |Coverage Status| image:: http://codecov.io/github/sdvillal/whatami/coverage.svg?branch=master :target: http://codecov.io/github/sdvillal/whatami?branch=master .. |Pypi Version| image:: https://badge.fury.io/py/whatami.svg :target: http://badge.fury.io/py/whatami .. _semantic versioning: http://semver.org/ .. _scikit-learn: http://scikit-learn.org .. _pyopy: https://github.com/sdvillal/pyopy .. |Codacy Status| image:: https://api.codacy.com/project/badge/Grade/163584371aeb432497b6c9525a21cdae :target: https://www.codacy.com/app/sdvillal/whatami?utm_source=github.com&utm_medium=referral&utm_content=sdvillal/whatami&utm_campaign=Badge_Grade