Change how the slice is stringified so it is easier to read.

From format of "FeatureA_X_FeatureB:ValueA_X_ValueB" to "FeautreA=ValueA; FeatureB=ValueB".
For fairness indicator, the original behavir was kept.

PiperOrigin-RevId: 446577668

RELEASE.md

@@ -11,6 +11,8 @@
 ## Bug Fixes and Other Changes
 
 ## Breaking Changes
+*   Changes how cross-slice is stringified for easier read: "FeatureA:ValueA__XX__FeatureB:ValueB" to "FeautreA=ValueA Vs.
+    FeatureB=ValueB".
 
 ## Deprecations
 

fairness_indicators/example_model_test.py

@@ -25,7 +25,6 @@
 import six
 import tensorflow.compat.v1 as tf
 import tensorflow_model_analysis as tfma
-from tensorflow_model_analysis.slicer import slicer_lib as slicer
 
 tf.compat.v1.enable_eager_execution()
 
@@ -96,7 +95,7 @@ def test_example_model(self):
                                  FEATURE_MAP)
 
     expected_slice_keys = [
-        'Overall', 'slice:slice3', 'slice:slice1', 'slice:slice2'
+        'Overall', 'slice: slice3', 'slice: slice1', 'slice: slice2'
     ]
     evaluation_results = tfma.load_eval_result(tfma_eval_result_path)
 
@@ -105,7 +104,7 @@ def test_example_model(self):
     # Verify if false_positive_rate metrics are computed for all values of
     # slice.
     for (slice_key, metric_value) in evaluation_results.slicing_metrics:
-      slice_key = slicer.stringify_slice_key(slice_key)
+      slice_key = tfma.slicer.stringify_slice_key(slice_key)
       self.assertIn(slice_key, expected_slice_keys)
       self.assertGreaterEqual(
           1.0, metric_value['']['']