def predict_boolq(self, payload):
start = time.time()
inp = {
"input_text": payload.get("input_text"),
"max_questions": payload.get("max_questions", 4)
}
text = inp['input_text']
num = inp['max_questions']
sentences = tokenize_sentences(text)
joiner = " "
modified_text = joiner.join(sentences)
answer = self.random_choice()
form = "truefalse: %s passage: %s </s>" % (modified_text, answer)
encoding = self.tokenizer.encode_plus(form, return_tensors="pt")
input_ids, attention_masks = encoding["input_ids"].to(
self.device), encoding["attention_mask"].to(self.device)
output = beam_search_decoding(input_ids, attention_masks, self.model,
self.tokenizer)
if torch.device == 'cuda':
torch.cuda.empty_cache()
final = {}
final['Text'] = text
final['Count'] = num
final['Boolean Questions'] = output
return final
def predict_mcq(self, payload):
start = time.time()
inp = {
"input_text": payload.get("input_text"),
"max_questions": payload.get("max_questions", 4)
}
text = inp['input_text']
sentences = tokenize_sentences(text)
joiner = " "
modified_text = joiner.join(sentences)
print(modified_text)
keywords = get_keywords(self.nlp, modified_text, inp['max_questions'],
self.s2v, self.fdist,
self.normalized_levenshtein, len(sentences))
print(keywords)
keyword_sentence_mapping = get_sentences_for_keyword(
keywords, sentences)
print(keyword_sentence_mapping)
for k in keyword_sentence_mapping.keys():
text_snippet = " ".join(keyword_sentence_mapping[k][:3])
keyword_sentence_mapping[k] = text_snippet
final_output = {}
if len(keyword_sentence_mapping.keys()) == 0:
return final_output
else:
try:
generated_questions = generate_questions_mcq(
keyword_sentence_mapping, self.device, self.tokenizer,
self.model, self.s2v, self.normalized_levenshtein)
except:
return final_output
end = time.time()
final_output["statement"] = modified_text
final_output["questions"] = generated_questions["questions"]
final_output["time_taken"] = end - start
if torch.device == 'cuda':
torch.cuda.empty_cache()
return final_output
def predict_shortq(self, payload):
inp = {
"input_text": payload.get("input_text"),
"max_questions": payload.get("max_questions", 4)
}
text = inp['input_text']
sentences = tokenize_sentences(text)
joiner = " "
modified_text = joiner.join(sentences)
keywords = get_keywords(self.nlp, modified_text, inp['max_questions'],
self.s2v, self.fdist,
self.normalized_levenshtein, len(sentences))
keyword_sentence_mapping = get_sentences_for_keyword(
keywords, sentences)
for k in keyword_sentence_mapping.keys():
text_snippet = " ".join(keyword_sentence_mapping[k][:3])
keyword_sentence_mapping[k] = text_snippet
final_output = {}
if len(keyword_sentence_mapping.keys()) == 0:
print('ZERO')
return final_output
else:
generated_questions = generate_normal_questions(
keyword_sentence_mapping, self.device, self.tokenizer,
self.model)
print(generated_questions)
final_output["statement"] = modified_text
final_output["questions"] = generated_questions["questions"]
if torch.device == 'cuda':
torch.cuda.empty_cache()
return final_output
