workflow patch

README.md

@@ -19,6 +19,9 @@ To install from source.
  3. Download the source of oxonfair and in the source directory run:
     pip install -e .
 
+To install every single dependency used for testing [autogluon, maplotlib, xgboost, and a range of helper functions] use:
+    pip install  -e . \[tests\]
+
 Now run the [Example Notebook](examples/quickstart_autogluon.ipynb) or try some of the example below.
 
 For scikit/XGBoost learn see [sklearn.md](./sklearn.md) and the [Example Notebook](examples/quickstart_xgboost.ipynb)

tests/unittests/test_ag.py

@@ -188,13 +188,13 @@ def test_recall_diff_inferred(use_fast=True):
     # difference between protected attributes of less than 2.5%) despite not
     # using sex at run-time
 
-    fpredictor.fit(gm.accuracy, gm.recall.diff, 0.025)
+    fpredictor.fit(gm.accuracy, gm.recall.diff, 0.001)
 
     measures = fpredictor.evaluate_fairness(verbose=False)
 
-    assert measures["original"]["recall.diff"] > 0.025
+    assert measures["original"]["recall.diff"] > 0.001
 
-    assert measures["updated"]["recall.diff"] < 0.025
+    assert measures["updated"]["recall.diff"] < 0.001
 
     # Prove that sex isn't being used by dropping it and reevaluating.