def predict(self, input):
prediction = list()
# Load lables
with open('./output_artefacts/{LABLE_TO_IX.JSON}') as json_file:
label_to_ix = json.load(json_file)
# Load config
with open('./output_artefacts/{CONFIG.PKL}', 'rb') as f:
config = pickle.load(f)
config.num_labels = len(list(label_to_ix.values()))
model = DistilBertForSequenceClassification(config)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# model = model.cuda()
model_path = './output_artefacts/{MODEL.PTH}'
model.load_state_dict(torch.load(model_path, map_location=device))
msg = preprocessing(str(input))
model.eval()
input_msg, _ = prepare_features(msg)
if torch.cuda.is_available():
input_msg = input_msg.cuda()
output = model(input_msg)[0]
output_exp = torch.exp(output)
probability = torch.div(output_exp, torch.add(output_exp, 1.))
_, pred_label = probability.topk(1)
percent = torch.mul(_, 100)
percent = torch.reshape(percent, (1, 1)).tolist()
percent = [item for sublist in percent for item in sublist]
for i in pred_label[0]:
queue = list(label_to_ix.keys())[i]
prediction.append(queue)
my_result = dict(zip(prediction, percent))
return my_result
def create_and_check_distilbert_for_sequence_classification(
self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels
):
config.num_labels = self.num_labels
model = DistilBertForSequenceClassification(config)
model.to(torch_device)
model.eval()
result = model(input_ids, attention_mask=input_mask, labels=sequence_labels)
self.parent.assertEqual(result.logits.shape, (self.batch_size, self.num_labels))
def create_and_check_distilbert_for_sequence_classification(self, config, input_ids, input_mask, sequence_labels, token_labels, choice_labels):
config.num_labels = self.num_labels
model = DistilBertForSequenceClassification(config)
model.eval()
loss, logits = model(input_ids, attention_mask=input_mask, labels=sequence_labels)
result = {
"loss": loss,
"logits": logits,
}
self.parent.assertListEqual(
list(result["logits"].size()),
[self.batch_size, self.num_labels])
self.check_loss_output(result)
