expan 0.6.9

Experiment Analysis Library

A/B tests (a.k.a. Randomized Controlled Trials or Experiments) have been widely applied in different industries to optimize business processes and user experience. ExpAn (Experiment Analysis) is a Python library developed for the statistical analysis of such experiments and to standardise the data structures used.

The data structures and functionality of ExpAn are generic such that they can be used by both data scientists optimizing a user interface and biologists running wet-lab experiments. The library is also standalone and can be imported and used from within other projects and from the command line.

Major statistical functionalities include:

• delta

• subgroup analysis

• trend analysis (coming soon)

Installation

Stable release

To install ExpAn, run this command in your terminal:

$ pip install expan

From sources

The sources for ExpAn can be downloaded from the Github repo.

You can either clone the public repository:

$ git clone git://github.com/zalando/expan

Or download the tarball:

$ curl  -OL https://github.com/zalando/expan/tarball/master

Once you have a copy of the source, you can install it with:

$ python setup.py install

Usage

To use ExpAn in a project:

import expan

Some mock-up data:

from expan.core.experiment import Experiment
from expan.core.util import generate_random_data

data, metadata = generate_random_data()
exp = Experiment(control_variant_name='B', data=data, metadata=metadata)
exp.delta()

Documentation

The latest stable version is 0.6.9.

ExpAn main documentation

License

The MIT License (MIT)

Copyright © [2016] Zalando SE, https://tech.zalando.com

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Change Log

v0.6.7 (2018-01-10)

Full Changelog

Closed issues:

• Group Sequential - Percentile Issue #176

Merged pull requests:

• Increase version to 0.6.7 #181 (shansfolder)

• fixed last command in “Deploying to PyPI” part of contributing.rst #180 (mkolarek)

• Extended multiple correction for group sequential, added doc for multiple correction. #179 (daryadedik)

• Fix information fraction calculation #177 (shansfolder)

v0.6.6 (2017-11-27)

Full Changelog

Closed issues:

• Infinitely large confidence intervals produced by group_sequential_delta() #172

Merged pull requests:

• Merging dev to master for new release #175 (mkolarek)

• AXO-35 implemented estimate_sample_size() for estimating sample size … #174 (mkolarek)

• Fix two-sided alpha value in power analysis #173 (shansfolder)

• Docs/update contrib doc #171 (mkolarek)

• Add some parameter checks #170 (shansfolder)

• Make applying bins to data frames more agreeable #169 (gbordyugov)

• OCTO-2181: Implement over time analysis. Time-based SGA #164 (daryadedik)

v0.6.5 (2017-10-24)

Full Changelog

Merged pull requests:

• updated version #168 (mkolarek)

• Bump version: 0.6.3 → 0.6.4 #167 (mkolarek)

• bump version to v0.6.3 #166 (mkolarek)

v0.6.3 (2017-10-24)

Full Changelog

Merged pull requests:

• OCTO-2214 Bugfix: Capping information fraction #163 (shansfolder)

• OCTO-2088: Implement multiple testing correction in ExpAn #161 (daryadedik)

• OCTO-1044 Improve readthedoc #160 (shansfolder)

• OCTO-1933 Subgroup analysis #159 (shansfolder)

• release 0.6.2 #156 (mkolarek)

• OCTO-1920, OCTO-1968, OCTO-1969 Refactor binning #155 (shansfolder)

v0.6.2 (2017-08-29)

Full Changelog

Fixed bugs:

• Result statistics in Baeysian methods #142

Closed issues:

• Default Parameters of Constructor of Experiment class #151

• Update to ExpAn-Intro.ipynb #141

Merged pull requests:

• make development requirements open ended #154 (mkolarek)

• Octo 1930 implement quantile filtering #153 (mkolarek)

• Not use empty list for method parameter #152 (shansfolder)

• OCTO-1971 Add variational inference for early stopping #150 (shansfolder)

• Updated intro documentation covering delta methods. #149 (daryadedik)

• Release v0.6.1 #148 (shansfolder)

• Merge pull request #137 from zalando/dev #147 (shansfolder)

• Add static html file from intro doc for v0.6.1 #146 (shansfolder)

v0.6.1 (2017-08-08)

Full Changelog

Implemented enhancements:

• Optimizing the control flow from Experiment to Results #82

• more meaningful dict keys for results #139 (gbordyugov)

Fixed bugs:

• reenable means and bounds functions on Results object #9

Closed issues:

• Results.to_json() implementation not flexible #65

• Results.to_json() doesn’t support trend() results #64

Merged pull requests:

• Documentation updates for Expan 0.6.x. Covers OCTO-1961, OCTO-1970 #145 (daryadedik)

• Fix delta/alpha model para inconsistency #144 (shansfolder)

• Small improvement on default type of report_kpi_names #140 (shansfolder)

• slightly different json structure for results #138 (gbordyugov)

• merging dev to master #137 (gbordyugov)

v0.6.0 (2017-07-26)

Full Changelog

Closed issues:

• Improve binning performance #135

• Missing unit tests for to_json() on early stopping algos #128

Merged pull requests:

• Octo 1616 no experimentdata #134 (gbordyugov)

• Attempt to fix pickling bug #133 (shansfolder)

• Stan models compilation, exceptions catch, unit tests adaptation. #131 (daryadedik)

• Added try-finally block for the compulsory clean-up of .pkl compiled models #130 (daryadedik)

• OCTO-1837 fixed to_json() #129 (gbordyugov)

v0.5.3 (2017-06-26)

Full Changelog

Implemented enhancements:

• Weighted KPIs is only implemented in regular delta #114

Fixed bugs:

• Assumption of nan when computing weighted KPIs #119

• Weighted KPIs is only implemented in regular delta #114

• Percentiles value is lost during computing group_sequential_delta #108

Closed issues:

• Failing early stopping unit tests #85

Merged pull requests:

• Release new version 0.5.3 #127 (mkolarek)

• OCTO-1804: Optimize the loading of .stan model in expan. #126 (daryadedik)

• Test travis python version #125 (shansfolder)

• OCTO-1619 Cleanup ExpAn code #124 (shansfolder)

• OCTO-1748: Make number of iterations as a method argument in _bayes_sampling #123 (daryadedik)

• OCTO-1615 Use Python builtin logging instead of our own debugging.py #122 (shansfolder)

• OCTO-1711 Support weighted KPIs in early stopping #121 (shansfolder)

• Fixed a few bugs #120 (shansfolder)

• OCTO-1614 cleanup module structure #115 (shansfolder)

• OCTO-1677 : fix missing .stan files #113 (gbordyugov)

• Bump version 0.5.1 -> 0.5.2 #112 (mkolarek)

v0.5.2 (2017-05-11)

Full Changelog

Implemented enhancements:

• OCTO-1502: cleanup of call chains #110 (gbordyugov)

Merged pull requests:

• OCTO-1502 support **kwargs for four delta functions #111 (shansfolder)

• new version 0.5.1 #107 (mkolarek)

v0.5.1 (2017-04-20)

Full Changelog

Implemented enhancements:

• Derived KPIs are passed to Experiment.fixed_horizon_delta() but never used in there #96

Merged pull requests:

• updated CONTRIBUTING.rst with deployment flow #106 (mkolarek)

• OCTO-1501: bugfix in Results.to_json() #105 (gbordyugov)

• OCTO-1502 removed variant_subset parameter… #104 (gbordyugov)

• OCTO-1540 cleanup handling of derived kpis #102 (shansfolder)

• OCTO-1540: cleanup of derived kpi handling in Experiment.delta() and … #97 (gbordyugov)

• Small refactoring #95 (shansfolder)

• Merge dev to master for v0.5.0 #94 (mkolarek)

v0.5.0 (2017-04-05)

Full Changelog

Implemented enhancements:

• Bad code duplication in experiment.py #81

• pip == 8.1.0 requirement #76

Fixed bugs:

• Experiment.sga() assumes features and KPIs are merged in self.metrics #87

• pctile can be undefined in Results.to\_json\(\) #78

Closed issues:

• Results.to_json() => TypeError: Object of type ‘UserWarning’ is not JSON serializable #77

• Rethink Results structure #66

Merged pull requests:

• new dataframe tree traverser in to_json() #92 (gbordyugov)

• updated requirements.txt to have ‘greater than’ dependencies instead … #89 (mkolarek)

• pip version requirement #88 (gbordyugov)

• Test #86 (s4826)

• merging in categorical binning #84 (gbordyugov)

• Add documentation of the weighting logic #83 (jbao)

• Early stopping #80 (jbao)

• a couple of minor cleanups #79 (gbordyugov)

• Merge to_json() changes #75 (mkolarek)

• Feature/early stopping #73 (jbao)

v0.4.5 (2017-02-10)

Full Changelog

Fixed bugs:

• Numbers cannot appear in variable names for derived metrics #58

Merged pull requests:

• Feature/results and to json refactor #74 (mkolarek)

• Merge to_json() and prob_uplift_over_zero changes #72 (mkolarek)

• regex fix, see https://github.com/zalando/expan/issues/58 #70 (gbordyugov)

v0.4.4 (2017-02-09)

Full Changelog

Implemented enhancements:

• Add argument assume_normal and treatment_cost to calculate_prob_uplift_over_zero() and prob_uplift_over_zero_single_metric() #26

• host intro slides (from the ipython notebook) somewhere for public viewing #10

Closed issues:

• migrate issues from github enterprise #20

Merged pull requests:

• Feature/results and to json refactor #71 (mkolarek)

• new to_json() functionality and improved vim support #67 (mkolarek)

v0.4.3 (2017-02-07)

Full Changelog

Closed issues:

• coverage % is misleading #23

Merged pull requests:

• Vim modelines #63 (gbordyugov)

• Feature/octo 1253 expan results in json #62 (mkolarek)

• 0.4.2 release #60 (mkolarek)

v0.4.2 (2016-12-08)

Full Changelog

Fixed bugs:

• frequency table in the chi square test doesn’t respect the order of categories #56

Merged pull requests:

• OCTO-1143 Review outlier filtering #59 (domheger)

• Workaround to fix #56 #57 (jbao)

v0.4.1 (2016-10-18)

Full Changelog

Merged pull requests:

• small doc cleanup #55 (jbao)

• Add comments to cli.py #54 (igusher)

• Feature/octo 545 add consolidate documentation #53 (mkolarek)

• added os.path.join instead of manual string concatenations with ‘/’ #52 (mkolarek)

• Feature/octo 958 outlier filtering #50 (mkolarek)

• Sort KPIs in reverse order before matching them in the formula #49 (jbao)

v0.4.0 (2016-08-19)

Full Changelog

Closed issues:

• Support ‘overall ratio’ metrics (e.g. conversion rate/return rate) as opposed to per-entity ratios #44

Merged pull requests:

• merging dev to master #48 (jbao)

• OCTO-825 overall metric #47 (jbao)

• Bump version: 0.3.2 → 0.3.4 #46 (mkolarek)

• Bug/fix dependencies #45 (mkolarek)

v0.3.4 (2016-08-08)

Full Changelog

Closed issues:

• perform trend analysis cumulatively #31

• Python3 #21

Merged pull requests:

• Feature/2to3 #43 (mkolarek)

v0.3.3 (2016-08-02)

Full Changelog

Merged pull requests:

• Merge pull request #41 from zalando/master #42 (jbao)

• master to dev #41 (mkolarek)

• Bump version: 0.3.1 → 0.3.2 #40 (mkolarek)

• Revert “Merge pull request #35 from zalando/dev” #39 (mkolarek)

• Merge pull request #35 from zalando/dev #38 (mkolarek)

v0.3.2 (2016-08-02)

Full Changelog

Merged pull requests:

• Bugfix/trend analysis bin label #37 (jbao)

• Added cumulative trends analysis OCTO-814 #36 (domheger)

• Merging 0.3.1 to master #35 (domheger)

v0.3.1 (2016-07-15)

Full Changelog

Merged pull requests:

• Bugfix/prob uplift over 0 #34 (jbao)

• Master #30 (mkolarek)

v0.3.0 (2016-06-23)

Full Changelog

Implemented enhancements:

• Add P(uplift>0) as a statistic #2

• Added function to calculate P(uplift>0) #24 (jbao)

Merged pull requests:

• updated travis.yml #29 (mkolarek)

• Master #28 (mkolarek)

• Master #27 (mkolarek)

• only store the p-value in the chi-square test result object #22 (jbao)

v0.2.5 (2016-05-30)

Full Changelog

Implemented enhancements:

• Implement __version__ #14

Closed issues:

• upload full documentation! #1

Merged pull requests:

• implement expan.__version__ #19 (pangeran-bottor)

• Mainly documentation changes, as well as travis config updates #17 (robertmuil)

• Update README.rst #16 (pangeran-bottor)

• added cli module #11 (mkolarek)

• new travis config specifying that only master and dev should be built #4 (mkolarek)

v0.2.4 (2016-05-16)

Full Changelog

Closed issues:

• No module named experiment and test_data #13

Merged pull requests:

• new travis config specifying that only master and dev should be built #5 (mkolarek)

v0.2.3 (2016-05-06)

Full Changelog

v0.2.2 (2016-05-06)

Full Changelog

v0.2.1 (2016-05-06)

Full Changelog

v0.2.0 (2016-05-06)

Merged pull requests:

• Added detailed documentation with data formats #3 (robertmuil)

* This Change Log was automatically generated by `github_changelog_generator <https://github.com/skywinder/Github-Changelog-Generator>`__