-
Notifications
You must be signed in to change notification settings - Fork 1
/
nlu.py
53 lines (39 loc) · 2.13 KB
/
nlu.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
import numpy as np
from tensorflow import keras
from keras.preprocessing.sequence import pad_sequences
from keras.layers import Bidirectional
import pickle
from tensorflow.keras.models import load_model
# Loading model files, getting initial NLU classes and filtering them within scope
ls = ['are_you_a_bot','calendar','calendar_update','cancel','change_volume','date','definition','flip_coin','fun_fact','how_old_are_you','meaning_of_life','next_holiday','next_song','play_music','reminder','reminder_update','repeat','roll_dice','spelling','tell_joke','time','timer','what_are_your_hobbies','what_can_i_ask_you','what_is_your_name','what_song','where_are_you_from','who_do_you_work_for','who_made_you']
class IntentClassifier:
def __init__(self,classes,model,tokenizer,label_encoder):
self.classes = classes
self.classifier = model
self.tokenizer = tokenizer
self.label_encoder = label_encoder
def get_intent(self,text):
self.text = [text]
self.test_keras = self.tokenizer.texts_to_sequences(self.text)
self.test_keras_sequence = pad_sequences(self.test_keras, maxlen=16, padding='post')
self.pred = self.classifier.predict(self.test_keras_sequence)
self.ls_out = []
self.ls_pred = np.argsort(np.max(self.pred[:],axis=0))[-3:]
for item in self.ls_pred:
if self.label_encoder.inverse_transform([item])[0] in ls:
self.ls_out.append(self.label_encoder.inverse_transform([item])[0])
else:
if 'oos' not in self.ls_out:
self.ls_out.append('oos')
return self.ls_out[::-1]
model = load_model('models/intents.h5', custom_objects={'Bidirectional': Bidirectional},compile=False)
with open('utils/classes.pkl','rb') as file:
classes = pickle.load(file)
with open('utils/tokenizer.pkl','rb') as file:
tokenizer = pickle.load(file)
with open('utils/label_encoder.pkl','rb') as file:
label_encoder = pickle.load(file)
def infer_intent(text):
nlu = IntentClassifier(classes,model,tokenizer,label_encoder)
intent = nlu.get_intent(text)
return intent