Update utils/vulnerability_classifier.py
Browse files
utils/vulnerability_classifier.py
CHANGED
@@ -9,7 +9,51 @@ import streamlit as st
|
|
9 |
from transformers import pipeline
|
10 |
from setfit import SetFitModel
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
13 |
@st.cache_resource
|
14 |
def load_vulnerabilityClassifier(config_file:str = None, classifier_name:str = None):
|
15 |
"""
|
@@ -89,12 +133,11 @@ def vulnerability_classification(haystack_doc:pd.DataFrame,
|
|
89 |
|
90 |
predictions = classifier_model(list(haystack_doc.text))
|
91 |
|
92 |
-
|
93 |
|
94 |
-
|
95 |
-
# for i in range(len(predictions)):
|
96 |
|
97 |
-
|
98 |
# placeholder = {}
|
99 |
# for j in range(len(temp)):
|
100 |
# placeholder[temp[j]['label']] = temp[j]['score']
|
|
|
9 |
from transformers import pipeline
|
10 |
from setfit import SetFitModel
|
11 |
|
12 |
+
label_dict= {0: 'Agricultural communities',
|
13 |
+
1: 'Children',
|
14 |
+
2: 'Coastal communities',
|
15 |
+
3: 'Ethnic, racial or other minorities',
|
16 |
+
4: 'Fishery communities',
|
17 |
+
5: 'Informal sector workers',
|
18 |
+
6: 'Members of indigenous and local communities',
|
19 |
+
7: 'Migrants and displaced persons',
|
20 |
+
8: 'Older persons',
|
21 |
+
9: 'Other',
|
22 |
+
10: 'Persons living in poverty',
|
23 |
+
11: 'Persons with disabilities',
|
24 |
+
12: 'Persons with pre-existing health conditions',
|
25 |
+
13: 'Residents of drought-prone regions',
|
26 |
+
14: 'Rural populations',
|
27 |
+
15: 'Sexual minorities (LGBTQI+)',
|
28 |
+
16: 'Urban populations',
|
29 |
+
17: 'Women and other genders'}
|
30 |
|
31 |
+
def getlabels(preds):
|
32 |
+
# Get label names
|
33 |
+
preds_list = preds.tolist()
|
34 |
+
|
35 |
+
predictions_names=[]
|
36 |
+
|
37 |
+
# loop through each prediction
|
38 |
+
for ele in preds_list:
|
39 |
+
|
40 |
+
# see if there is a value 1 and retrieve index
|
41 |
+
try:
|
42 |
+
index_of_one = ele.index(1)
|
43 |
+
except ValueError:
|
44 |
+
index_of_one = "NA"
|
45 |
+
|
46 |
+
# Retrieve the name of the label (if no prediction made = NA)
|
47 |
+
if index_of_one != "NA":
|
48 |
+
name = label_dict[index_of_one]
|
49 |
+
else:
|
50 |
+
name = "NA"
|
51 |
+
|
52 |
+
# Append name to list
|
53 |
+
predictions_names.append(name)
|
54 |
+
|
55 |
+
return predictions_names
|
56 |
+
|
57 |
@st.cache_resource
|
58 |
def load_vulnerabilityClassifier(config_file:str = None, classifier_name:str = None):
|
59 |
"""
|
|
|
133 |
|
134 |
predictions = classifier_model(list(haystack_doc.text))
|
135 |
|
136 |
+
|
137 |
|
138 |
+
pred_labels = getlabels(predictions)
|
|
|
139 |
|
140 |
+
haystack_doc['Vulnerability Label'] = pred_labels
|
141 |
# placeholder = {}
|
142 |
# for j in range(len(temp)):
|
143 |
# placeholder[temp[j]['label']] = temp[j]['score']
|