Fix/anchor base coverage update (#919)

* Fixed AnchorBased coverage updated for proposed anchors after computation.

* Removed commented lines and removed unecessary dummy updates of coverage from tests.

Author: RobertSamoilescu

alibi/explainers/anchors/anchor_base.py

@@ -458,8 +458,6 @@ def update_state(self, covered_true: np.ndarray, covered_false: np.ndarray, labe
         self.state['t_idx'][anchor].update(idxs)
         self.state['t_nsamples'][anchor] += n_samples
         self.state['t_positives'][anchor] += labels.sum()
-        if coverage > -1:
-            self.state['t_coverage'][anchor] = coverage
         self.state['t_covered_true'][anchor] = covered_true
         self.state['t_covered_false'][anchor] = covered_false
         self.state['data'][idxs] = data

alibi/explainers/tests/test_anchor_base.py

@@ -16,7 +16,6 @@ def test_anchor_base_beam(rf_classifier, at_defaults, at_iris_explainer):
     # inputs
     n_anchors_to_sample = 6
     coverage_samples = 500
-    dummy_coverage = - 0.55  # used to test coverage updates on sampling
 
     X_test, explainer, predict_fn, predict_type = at_iris_explainer
     explain_defaults = at_defaults
@@ -37,15 +36,10 @@ def test_anchor_base_beam(rf_classifier, at_defaults, at_iris_explainer):
         to_sample.append(tuple(anchor))
     to_sample = list(set(to_sample))
     current_state = deepcopy(anchor_beam.state)
-    for anchor in to_sample:
-        if anchor not in current_state['t_nsamples']:
-            anchor_beam.state['t_coverage'][anchor] = dummy_coverage
     pos, total = anchor_beam.draw_samples(to_sample, explain_defaults['batch_size'])
     for p, t, anchor in zip(pos, total, to_sample):
         assert anchor_beam.state['t_nsamples'][anchor] == current_state['t_nsamples'][anchor] + t
         assert anchor_beam.state['t_positives'][anchor] == current_state['t_positives'][anchor] + p
-        if anchor:  # empty anchor has dummy coverage
-            assert anchor_beam.state['t_coverage'][anchor] != dummy_coverage
 
     # testing resampling works
     # by sampling all features, we are guaranteed that partial anchors might not exist