def test_set_custom_classifier(self):
all_data = get_ssts_dataset('train')
model_config = RNN_MODEL_CONFIG
preprocessor = GlovePreprocessor(TOKEN_COUNT_THRESHOLD, MAX_SENT_COUNT)
preprocessor.build_vocab(all_data[TEXT_COL])
explainer = IntrospectiveRationaleExplainer(
classifier_type=CLASSIFIER_TYPE_CUSTOM, cuda=CUDA)
explainer.build_model_config(model_config)
# set custom classifier
classifier = ClassifierModule(explainer.get_model_config(),
preprocessor.word_vocab)
explainer.set_preprocessor(preprocessor)
explainer.set_classifier(classifier)
assert explainer.classifier_type is CLASSIFIER_TYPE_CUSTOM
def set_classifier(self,
classifier_type=CLASSIFIER_TYPE_RNN,
classifier=None):
""" Set the classifier from prepackaged option or provide custom classifier
:param classifier_type: One of ['BERT', 'RNN', 'BERT_RNN']
:type: str
:param classifier: Custom provided classifier
:type: Any
"""
if classifier_type == CLASSIFIER_TYPE_RNN or classifier_type == CLASSIFIER_TYPE_BERT_RNN:
self.classifier = ClassifierModule(self.model_config,
self.preprocessor.word_vocab)
elif classifier_type == CLASSIFIER_TYPE_BERT:
self.classifier = BertForSequenceClassification.from_pretrained(
"bert-base-uncased",
num_labels=self.model_config.num_labels,
output_hidden_states=False,
output_attentions=False,
)
else:
self.classifier = classifier
def set_generator_classifier(self,
classifier_type=CLASSIFIER_TYPE_RNN,
generator_classifier=None):
""" Set classifier for the Generator
:param classifier_type: One of ['BERT', 'RNN', 'BERT_RNN']
:type classifier_type: str
:param generator_classifier: Custom provided classifier for generator
:type generator_classifier: Any
:return: Any
"""
if classifier_type == CLASSIFIER_TYPE_RNN:
self.generator_classifier = ClassifierModule(
self.model_config, self.preprocessor.word_vocab)
elif classifier_type == CLASSIFIER_TYPE_BERT or classifier_type == CLASSIFIER_TYPE_BERT_RNN:
self.generator_classifier = BertForSequenceClassification.from_pretrained(
"bert-base-uncased",
num_labels=self.model_config.num_labels,
output_hidden_states=True,
output_attentions=True,
)
else:
self.generator_classifier = generator_classifier