def classify_string_classifier_rule(state, normalized_answer):
"""Run the classifier if no prediction has been made yet. Currently this
is behind a development flag.
"""
best_matched_answer_group = None
best_matched_answer_group_index = len(state.interaction.answer_groups)
best_matched_rule_spec_index = None
best_matched_truth_value = rule_domain.CERTAIN_FALSE_VALUE
sc = classifier_services.StringClassifier()
training_examples = [
[doc, []] for doc in state.interaction.confirmed_unclassified_answers
]
for (answer_group_index,
answer_group) in enumerate(state.interaction.answer_groups):
fuzzy_rule_spec_index = answer_group.get_fuzzy_rule_index()
if fuzzy_rule_spec_index is not None:
fuzzy_rule_spec = answer_group.rule_specs[fuzzy_rule_spec_index]
else:
fuzzy_rule_spec = None
if fuzzy_rule_spec is not None:
training_examples.extend(
[[doc, [str(answer_group_index)]]
for doc in fuzzy_rule_spec.inputs['training_data']])
if len(training_examples) > 0:
sc.load_examples(training_examples)
doc_ids = sc.add_examples_for_predicting([normalized_answer])
predicted_label = sc.predict_label_for_doc(doc_ids[0])
if (predicted_label !=
classifier_services.StringClassifier.DEFAULT_LABEL):
predicted_answer_group_index = int(predicted_label)
predicted_answer_group = state.interaction.answer_groups[
predicted_answer_group_index]
best_matched_truth_value = rule_domain.CERTAIN_TRUE_VALUE
for rule_spec in predicted_answer_group.rule_specs:
if rule_spec.rule_type == rule_domain.FUZZY_RULE_TYPE:
best_matched_rule_spec_index = fuzzy_rule_spec_index
break
best_matched_answer_group = predicted_answer_group
best_matched_answer_group_index = predicted_answer_group_index
return {
'outcome': best_matched_answer_group.outcome.to_dict(),
'answer_group_index': best_matched_answer_group_index,
'classification_certainty': best_matched_truth_value,
'rule_spec_index': best_matched_rule_spec_index,
}
else:
return None
return None
def classify_string_classifier_rule(state, normalized_answer):
"""Run the classifier if no prediction has been made yet. Currently this
is behind a development flag.
"""
best_matched_answer_group = None
best_matched_answer_group_index = len(state.interaction.answer_groups)
best_matched_rule_spec_index = None
sc = classifier_services.StringClassifier()
training_examples = [
[doc, []] for doc in state.interaction.confirmed_unclassified_answers
]
for (answer_group_index,
answer_group) in enumerate(state.interaction.answer_groups):
classifier_rule_spec_index = answer_group.get_classifier_rule_index()
if classifier_rule_spec_index is not None:
classifier_rule_spec = answer_group.rule_specs[
classifier_rule_spec_index]
else:
classifier_rule_spec = None
if classifier_rule_spec is not None:
training_examples.extend(
[[doc, [str(answer_group_index)]]
for doc in classifier_rule_spec.inputs['training_data']])
if len(training_examples) > 0:
sc.load_examples(training_examples)
doc_ids = sc.add_docs_for_predicting([normalized_answer])
predicted_label = sc.predict_label_for_doc(doc_ids[0])
if (predicted_label !=
classifier_services.StringClassifier.DEFAULT_LABEL):
predicted_answer_group_index = int(predicted_label)
predicted_answer_group = state.interaction.answer_groups[
predicted_answer_group_index]
for rule_spec in predicted_answer_group.rule_specs:
if rule_spec.rule_type == exp_domain.CLASSIFIER_RULESPEC_STR:
best_matched_rule_spec_index = classifier_rule_spec_index
break
best_matched_answer_group = predicted_answer_group
best_matched_answer_group_index = predicted_answer_group_index
return {
'outcome': best_matched_answer_group.outcome.to_dict(),
'answer_group_index': best_matched_answer_group_index,
'rule_spec_index': best_matched_rule_spec_index,
}
else:
return None
return None
def setUp(self):
super(StringClassifierUnitTests, self).setUp()
self.string_classifier = classifier_services.StringClassifier()
self.string_classifier.load_examples(self._EXAMPLES_TRAIN)