Spaces:
Sleeping
Sleeping
Upload utils.py
Browse files
utils.py
ADDED
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import re
|
2 |
+
import pickle
|
3 |
+
import numpy as np
|
4 |
+
import pandas as pd
|
5 |
+
|
6 |
+
|
7 |
+
tfidf = pickle.load(open('models/tfidf.sav', 'rb'))
|
8 |
+
svc_sentiment = pickle.load(open('models/sentiment_model.sav', 'rb'))
|
9 |
+
tfidf_sentiment = pickle.load(open('models/tfidf_sentiment.sav', 'rb'))
|
10 |
+
svc = pickle.load(open('models/svc_model.sav', 'rb'))
|
11 |
+
|
12 |
+
labels = [
|
13 |
+
'Product quality', 'Knowledge',
|
14 |
+
'Appointment', 'Service etiquette', 'Waiting time',
|
15 |
+
'Repair speed', 'Repair cost', 'Repair quality', 'Warranty',
|
16 |
+
'Product replacement', 'Loan sets']
|
17 |
+
|
18 |
+
sample_file = pd.read_csv('sample.csv').to_csv(index=False).encode('utf-8')
|
19 |
+
|
20 |
+
print('utils imported!')
|
21 |
+
|
22 |
+
def get_single_prediction(text):
|
23 |
+
|
24 |
+
# manipulate data into a format that we pass to our model
|
25 |
+
text = text.lower().strip() #lower case
|
26 |
+
|
27 |
+
# Vectorise text and store in new dataframe. Sentence vector = average of word vectors
|
28 |
+
text_vectors = tfidf.transform(list(text))
|
29 |
+
|
30 |
+
# Make topic predictions
|
31 |
+
results = svc.predict_proba(text_vectors).squeeze().round(2)
|
32 |
+
pred_prob = pd.DataFrame({'topic': labels, 'probability': results}).sort_values('probability', ascending=True)
|
33 |
+
|
34 |
+
# Make sentiment predictions
|
35 |
+
text_vectors_sentiment = tfidf_sentiment.transform(list(text))
|
36 |
+
|
37 |
+
results_sentiment = svc_sentiment.predict_proba(text_vectors).squeeze().round(2)
|
38 |
+
pred_prob_sentiment = pd.DataFrame({'sentiment': ['Negative', 'Positive'], 'probability': results_sentiment}).sort_values('probability', ascending=True)
|
39 |
+
|
40 |
+
|
41 |
+
return (pred_prob, pred_prob_sentiment)
|
42 |
+
|
43 |
+
def get_multiple_predictions(csv):
|
44 |
+
|
45 |
+
df = pd.read_csv(csv)
|
46 |
+
df.columns = ['sequence']
|
47 |
+
|
48 |
+
df['sequence_clean'] = df['sequence'].str.lower().str.strip()
|
49 |
+
|
50 |
+
# Remove rows with blank string
|
51 |
+
invalid = df[(pd.isna(df['sequence_clean'])) | (df['sequence_clean'] == '')]
|
52 |
+
invalid.drop(columns=['sequence_clean'], inplace=True)
|
53 |
+
|
54 |
+
# Drop rows with blank string
|
55 |
+
df.dropna(inplace=True)
|
56 |
+
df = df[df['sequence_clean'] != ''].reset_index(drop=True)
|
57 |
+
|
58 |
+
# Vectorise text and get topic predictions
|
59 |
+
text_vectors = tfidf.transform(df['sequence_clean'])
|
60 |
+
pred_results = pd.DataFrame(svc.predict(text_vectors), columns = labels)
|
61 |
+
|
62 |
+
# Vectorise text and get sentiment predictions
|
63 |
+
text_vectors_sentiment = tfidf_sentiment.transform(df['sequence_clean'])
|
64 |
+
pred_results_sentiment = pd.DataFrame(svc_sentiment.predict(text_vectors_sentiment), columns = ['sentiment'])
|
65 |
+
|
66 |
+
# Join back to original sequence
|
67 |
+
final_results = df.join(pred_results).join(pred_results_sentiment)
|
68 |
+
final_results['others'] = final_results[labels].max(axis=1)
|
69 |
+
final_results['others'] = final_results['others'].apply(lambda x: 1 if x == 0 else 0)
|
70 |
+
|
71 |
+
final_results.drop(columns=['sequence_clean'], inplace=True)
|
72 |
+
|
73 |
+
# Append invalid rows
|
74 |
+
if len(invalid) == 0:
|
75 |
+
return final_results.to_csv(index=False).encode('utf-8')
|
76 |
+
else:
|
77 |
+
return pd.concat([final_results, invalid]).reset_index(drop=True).to_csv(index=False).encode('utf-8')
|