Signif subset (#73)

* binary tp, fp, tn, fn cols for statistical significance

* IP: statistical significance notebook testing binary tp, fp, tn, fn cols method in style of get_crosstabs for easy integration

* Calculate significance of differences between ref group and all other groups for label, score, tp, tn, fp, fn

* mask and move significance methods to bias class

* ID ref groups for significance methods

* Asignificance calculation, predefinied ref group bias method

* Add significance calculation to all bias methods

* IP: significance indicator in plotting disparity method

* unmasked statistical significance signaling in treemaps (masked still has error)

* masked statistical significance signaling in treemaps complete for all disparity metrics

* Add method descriptions, dynamic significance asterisks in treemaps

* rm test statistical siginificance

* Layer check for siginificance columns in df before attempting treemap significance asterisks.

* remove hard-coded metrics to check from significance from majority bias calculation

* Delete encodings.xml

* progress on passing tests with min_metric designation

* begin min_metric bias attempt w/o multiple groupbys

* threshold cutoff exploration

* NaN error fixed on min_metric testing. To do: verify numpy version re: TypeError on numpy boolean subtract

* min_metric now possible with test 3 and test 4 csv

* fix for type error on numpy boolean subtract

* clean run of ipynb

* ensure Plot() can handle multiple models

* ensure Plot() can handle multiple models

* Update src/aequitas/bias.py

value error if non-existing column provided for significance calcs

Co-Authored-By: lorenh516 <loren.hinkson@gmail.com>

* review comments

* breaking up long lines so easier to read

* post-significance pull COMPAS refresh

* mulitmodel plotting

* docs

* temp changes

* multimodel plotting

* plot multimodel, grouping changes

* can subset significance with COMPAS dataset

* sync with master

* all or none or selected possible

* cli handling

* cli handling subset

* no set for subset

* one ore

* no label_score_ref

* now?

* reset_index in function columns

* address indexing issue

* fix slicing error

* slicing

* remove print stmt

* remove .loc in binary func calc

* .loc in adding binaries

* dont specify level in adding binaries

* checkout

* warnings

* warnings

* report change

* multimodel fcns, CLI warnings fixed

* updates for warnings, serve debugging

* fix deprecated yaml load()

* config plotting args

* no warnings, webapp works

* docstrings

* clean

* clean

* NaN case wtd avg

* last version

* remove multimodel testing ipynb

* Update per jesteria suggestion

Co-Authored-By: Jesse London <jesselondon@gmail.com>

* code review

* clear

* signif with label score by default

* check

* bias merge

* small tweak

* fix signifiance miscalc

* update COMPAS

* tested server, CLI, .ipynb

README.md

@@ -38,7 +38,7 @@ Install Aequitas using pip:
     git clone https://github.com/dssg/aequitas.git
     cd aequitas
     python setup.py install
-(be mindful of the python version you use to run setup.py)
+(Note: be mindful of the python version you use to run setup.py)
 
 You may then import the ``aequitas`` module from Python: