def transform(self, texts):
_check_st()
model = SentenceTransformer('bert-base-nli-mean-tokens')
X = np.array(model.encode(texts))
return X
def transform(self, texts):
try:
from sentence_transformers.SentenceTransformer import SentenceTransformer #noqa
except ImportError:
print("Error: install sentence_transformers package "
"(`pip install sentence_transformers`)"
" to use Sentence BERT.")
sys.exit(192)
model = SentenceTransformer('bert-base-nli-mean-tokens')
X = np.array(model.encode(texts))
return X
def __init__(self,
faq_path: str,
faq_data: dict = None,
model_path: str = None):
"""
Either mention the model path to previously saved model ,
or let it be , none
when model path is None , the model will be a transformer model, with roBERTa base
faq_name is the name of the faq , generated questions and answers , if it already exists , we will used the
processed questions and answers , otherwise , we have to create a new one
if faq name , has not been processed atleast once , you must provide faq_data
faq_data --> dict has two keys , question_to_label , and answer_to_label
{q2l : {} , a2l : {}}
1) question_to_label
2) answer_to_label
q2l = {"How are you doing " : 1 , "where are you ? ": 3}
a2l = {"I am fine" : 1 , "I am in India": 3}
faq_data = {"questiontolabel" : q2l , "answertolabel" : a2l}
question_to_labels is again a dictionary from questions : label(int) can have multiple questions for same label
answer_to_labels is a dictionary from answers to label : one label per answer (strict !!!)
"""
if (model_path == None):
model_path = 'roberta-base-nli-stsb-mean-tokens'
self.model = SentenceTransformer(model_path)
self.current_faq = None
self.faq_path = faq_path
self.question_to_label = {
} # contans all the augmented and orignal questions mapped to their labels
self.answer_to_label = {} # contains mapping form answer to labels
# current data is to be filled using the fit_FAQ function call
# it has 3 keys 1) embeddings (a np array) 2) labels 3) label_to_answer dict
if (self.check_faq_path()):
print("found preexisiting faq data , loading dicts from the same")
que_path = os.path.join(self.faq_path, "questions.pkl")
self.question_to_label = load_dict(que_path)
ans_path = os.path.join(self.faq_path, "answers.pkl")
self.answer_to_label = load_dict(ans_path)
else:
assert not faq_data is None, "Did not find and preexisting of {} so you must provide faq_data".format(
faq_data)
self.make_faq(faq_data)