[Pyton][Arena] Add Fairnes plot (#373)

* [Pyton][Arena] Add Fairnes plot

* [Pyton] Update NEWS.md

python/dalex/NEWS.md

@@ -22,6 +22,7 @@ These are summed up in ([#368](https://github.com/ModelOriented/DALEX/issues/368
 
 * added `geom='bars'` to `AggregateProfiles.plot` to force the categorical plot 
 * added `geom='roc'` and `geom='lift'` to `ModelPerformance.plot`
+* added Fairness plot to Arena
 
 #### other
 

python/dalex/dalex/arena/object.py

@@ -85,7 +85,8 @@ def __init__(self, precalculate=False, enable_attributes=True, enable_custom_par
             CeterisParibusContainer,
             BreakDownContainer,
             MetricsContainer,
-            ROCContainer
+            ROCContainer,
+            FairnessCheckContainer
         ]
         self.options = {}
         for plot in self.plots:

python/dalex/dalex/arena/plots/__init__.py

@@ -6,6 +6,7 @@
 from ._ceteris_paribus_container import CeterisParibusContainer
 from ._metrics_container import MetricsContainer
 from ._roc_container import ROCContainer
+from ._fairness_check_container import FairnessCheckContainer
 
 __all__ = [
     'ShapleyValuesContainer',
@@ -15,5 +16,6 @@
     'CeterisParibusContainer',
     'BreakDownContainer',
     'MetricsContainer',
-    'ROCContainer'
+    'ROCContainer',
+    'FairnessCheckContainer'
 ]

python/dalex/dalex/arena/plots/_fairness_check_container.py

@@ -0,0 +1,53 @@
+import pandas as pd
+import numpy as np
+from pandas.api.types import is_object_dtype
+from .._plot_container import PlotContainer
+from dalex.fairness._group_fairness import utils, checks
+
+def rm_nan(obj):
+    return { k: (None if np.isnan(obj[k]) or np.isinf(obj[k]) else obj[k]) for k in obj.keys() }
+
+class FairnessCheckContainer(PlotContainer):
+    info = {
+        'name': 'Fairness',
+        'plotType': 'Fairness',
+        'plotCategory': 'Dataset Level',
+        'requiredParams': ['model', 'variable']
+    }
+    options = {
+        'cutoffs': { 'default': [x / 100 for x in range(5, 100, 5)], 'desc': 'List of tested cutoff levels' },
+    }
+    def _fit(self, model, variable):
+        if not variable.variable in model.variables:
+            raise Exception('Variable is not a column of explainer')
+        exp = model.explainer
+        y_hat = exp.predict(exp.data) if exp.y_hat is None else exp.y_hat
+        protected = exp.data[variable.variable]
+        if exp.model_type != 'classification':
+            self.set_message('Fairness plot is only available for classificators')
+            return
+        if not is_object_dtype(protected):
+            self.set_message('Select categorical variable to check fairness')
+            return
+
+        output_df = None
+        for cutoff in self.arena.get_option(self.plot_type, 'cutoffs'):
+            cutoff_dict = checks.check_cutoff(protected, cutoff, False)
+            sub_confusion_matrix = utils.SubgroupConfusionMatrix(exp.y, y_hat, protected, cutoff_dict)
+            sub_confusion_matrix_metrics = utils.SubgroupConfusionMatrixMetrics(sub_confusion_matrix)
+            df = sub_confusion_matrix_metrics.to_vertical_DataFrame()
+            df['cutoff'] = cutoff
+            output_df = df if output_df is None else output_df.append(df)
+
+        output = {}
+        for (subgroup, x) in output_df.set_index('metric').groupby('subgroup'):
+            output[subgroup] = {}
+            for (cutoff, y) in x.groupby('cutoff'):
+                output[subgroup][cutoff] = rm_nan(y['score'].to_dict())
+
+        self.data = { 'subgroups': output }
+
+    def test_arena(arena):
+        if type(arena).__name__ != 'Arena' or type(arena).__module__ != 'dalex.arena.object':
+            raise Exception('Invalid Arena argument')
+        return next((True for model in arena.get_params('model') if model.explainer.model_type == 'classification'), False)

python/dalex/dalex/arena/server.py

@@ -65,7 +65,7 @@ def get_params(request):
     @app.route("/<string:plot_type>", methods=['GET'])
     def get_plot(plot_type):
         if plot_type == 'timestamp':
-            return {'timestamp': arena.timestamp * 1000}
+            return Response(json.dumps({'timestamp': arena.timestamp * 1000}, default=convert), content_type='application/json')
         elif plot_type == 'shutdown':
             if request.args.get('token') != shutdown_token:
                 abort(403)

python/dalex/test/test_arena_classification.py

@@ -54,7 +54,7 @@ def setUp(self):
 
         # This plots should be supported
         self.reference_plots = [ROCContainer, ShapleyValuesContainer, BreakDownContainer, CeterisParibusContainer,
-            FeatureImportanceContainer, PartialDependenceContainer, AccumulatedDependenceContainer, MetricsContainer]
+            FeatureImportanceContainer, PartialDependenceContainer, AccumulatedDependenceContainer, MetricsContainer, FairnessCheckContainer]
 
     def test_supported_plots(self):
         arena = dx.Arena()