def run(self):
"""
Runs ER at all thresholds
:return predicted_labels: List of lists of predicted labels.
predicted_labels[threshold_index] = dict [identifier, cluster label]
:return metrics: List of lists of metric objects.
metrics[threshold_index] = Metrics object
:return er_objects: List of EntityResolution objects.
er_objects[threshold_index] = EntityResolution
:return new_metrics_objects: List of NewMetrics objects.
new_metrics_objects[threshold_index] = NewMetrics
"""
er = EntityResolution()
#weak_match_function = LogisticMatchFunction(self._database_train, self._labels_train, self._train_pair_seed, 0.5)
weak_match_function = ForestMatchFunction(self._database_train, self._labels_train, self._train_pair_seed, 0.5)
print 'Testing pairwise match function on test database'
ROC = weak_match_function.test(self._database_validation, self._labels_validation, self._validation_seed)
#ROC.make_plot()
metrics_list = list()
labels_list = list()
new_metrics_list = list()
class_balance_test = count_pairwise_class_balance(self._labels_test)
blocks = BlockingScheme(self._database_test, single_block=True)
for threshold in self.thresholds:
print 'Running entity resolution at threshold =', threshold
weak_match_function.set_decision_threshold(threshold)
labels_pred = weak_connected_components(self._database_test, weak_match_function, blocks)
#labels_pred = er.run(self._database_test, weak_match_function, single_block=True, max_block_size=np.Inf,
# cores=1)
metrics_list.append(Metrics(self._labels_test, labels_pred))
new_metrics_list.append(NewMetrics(self._database_test, labels_pred, weak_match_function, class_balance_test))
labels_list.append(labels_pred)
return labels_list, metrics_list, new_metrics_list
