Spaces:
Runtime error
Runtime error
sagittariusA
commited on
Commit
•
967e26e
1
Parent(s):
9520738
add application file
Browse files
app.py
ADDED
@@ -0,0 +1,79 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import torch.nn.functional as F
|
4 |
+
import numpy as np
|
5 |
+
from corpy.morphodita import Tokenizer
|
6 |
+
|
7 |
+
import transformers
|
8 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
9 |
+
|
10 |
+
model_checkpoint = 'ufal/robeczech-base'
|
11 |
+
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
|
12 |
+
transformers.logging.set_verbosity(transformers.logging.ERROR)
|
13 |
+
|
14 |
+
def classify_sentence(sent:str):
|
15 |
+
toksentence = tokenizer(sent,truncation=True,return_tensors="pt")
|
16 |
+
model.eval()
|
17 |
+
with torch.no_grad():
|
18 |
+
toksentence.to(device)
|
19 |
+
output = model(**toksentence)
|
20 |
+
|
21 |
+
return F.softmax(output.logits,dim=1).argmax(dim=1)
|
22 |
+
|
23 |
+
def classify_text(text:str):
|
24 |
+
tokenizer_morphodita = Tokenizer("czech")
|
25 |
+
|
26 |
+
all = []
|
27 |
+
for sentence in tokenizer_morphodita.tokenize(text, sents=True):
|
28 |
+
all.append(sentence)
|
29 |
+
|
30 |
+
sentences = np.array([' '.join(x) for x in all])
|
31 |
+
annotations = np.array(list(map(classify_sentence,sentences)))
|
32 |
+
|
33 |
+
return annotations
|
34 |
+
|
35 |
+
def classify_text_wrapper(text:str):
|
36 |
+
result = classify_text(text)
|
37 |
+
n = len(result)
|
38 |
+
non_biased = np.where(result==0)[0].shape[0]
|
39 |
+
biased = np.where(result==1)[0].shape[0]
|
40 |
+
|
41 |
+
return {'Non-biased':non_biased/n,'Biased':biased/n}
|
42 |
+
|
43 |
+
|
44 |
+
def interpret_bias(text:str):
|
45 |
+
result = classify_text(text)
|
46 |
+
|
47 |
+
tokenizer_morphodita = Tokenizer("czech")
|
48 |
+
|
49 |
+
interpretation = []
|
50 |
+
all = []
|
51 |
+
for sentence in tokenizer_morphodita.tokenize(text, sents=True):
|
52 |
+
all.append(sentence)
|
53 |
+
|
54 |
+
sentences = np.array([' '.join(x) for x in all])
|
55 |
+
|
56 |
+
for idx,sentence in enumerate(sentences):
|
57 |
+
score = 0
|
58 |
+
#non biased
|
59 |
+
if result[idx] == 0:
|
60 |
+
score = -1
|
61 |
+
#biased
|
62 |
+
if result[idx] == 1:
|
63 |
+
score = 1
|
64 |
+
interpretation.append((sentence, score))
|
65 |
+
|
66 |
+
return interpretation
|
67 |
+
|
68 |
+
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)
|
69 |
+
model = AutoModelForSequenceClassification.from_pretrained("sagittariusA/media_bias_classifier_cs")
|
70 |
+
model.eval()
|
71 |
+
|
72 |
+
label = gr.outputs.Label(num_top_classes=2)
|
73 |
+
inputs = gr.inputs.Textbox(placeholder=None, default="", label=None)
|
74 |
+
app = gr.Interface(fn=classify_text_wrapper,title='Bias classifier',theme='default',
|
75 |
+
inputs="textbox",layout='unaligned', outputs=label, capture_session=True
|
76 |
+
,interpretation=interpret_bias)
|
77 |
+
|
78 |
+
app.launch(inbrowser=True)
|
79 |
+
|