Spaces:
Runtime error
Runtime error
daniild71r
commited on
Commit
•
b9f69ed
1
Parent(s):
f37dabc
app created
Browse files- app.py +90 -2
- cat_mapping.json +42 -0
- cat_name_mapping.json +42 -0
- final_model/config.json +109 -0
- final_model/pytorch_model.bin +3 -0
- requirements.txt +2 -0
app.py
CHANGED
@@ -1,4 +1,92 @@
|
|
|
|
1 |
import streamlit as st
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
import streamlit as st
|
3 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification, TextClassificationPipeline
|
4 |
+
from tokenizers import Tokenizer
|
5 |
|
6 |
+
|
7 |
+
def fake_hash(x):
|
8 |
+
return 0
|
9 |
+
|
10 |
+
|
11 |
+
@st.cache(hash_funcs={Tokenizer: fake_hash}, suppress_st_warning=True, allow_output_mutation=True)
|
12 |
+
def initialize():
|
13 |
+
model_name = 'distilbert-base-cased'
|
14 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
15 |
+
model = AutoModelForSequenceClassification.from_pretrained('./final_model')
|
16 |
+
|
17 |
+
the_pipeline = TextClassificationPipeline(
|
18 |
+
model=model,
|
19 |
+
tokenizer=tokenizer,
|
20 |
+
return_all_scores=True,
|
21 |
+
device=-1
|
22 |
+
)
|
23 |
+
|
24 |
+
cat_mapping_file = open('cat_mapping.json', 'r')
|
25 |
+
cat_name_mapping_file = open('cat_name_mapping.json', 'r')
|
26 |
+
cat_mapping = json.load(cat_mapping_file)
|
27 |
+
cat_name_mapping = json.load(cat_name_mapping_file)
|
28 |
+
|
29 |
+
return the_pipeline, cat_mapping, cat_name_mapping
|
30 |
+
|
31 |
+
|
32 |
+
def get_top(the_pipeline, cat_mapping, title, summary, thresh=0.95):
|
33 |
+
if title == '' or summary == '':
|
34 |
+
return 'Not enough data to compute.'
|
35 |
+
|
36 |
+
question = title + ' || ' + summary
|
37 |
+
if len(question) > 4000:
|
38 |
+
return 'Your input is supsiciously long, try something shorter.'
|
39 |
+
|
40 |
+
try:
|
41 |
+
result = the_pipeline(question)[0]
|
42 |
+
result.sort(key=lambda x: -x['score'])
|
43 |
+
|
44 |
+
current_sum = 0
|
45 |
+
scores = []
|
46 |
+
|
47 |
+
for score in result:
|
48 |
+
scores.append(score)
|
49 |
+
current_sum += score['score']
|
50 |
+
if current_sum >= thresh:
|
51 |
+
break
|
52 |
+
|
53 |
+
for i in range(len(result)):
|
54 |
+
result[i]['label'] = cat_mapping[result[i]['label'][6:]]
|
55 |
+
|
56 |
+
return scores
|
57 |
+
|
58 |
+
except BaseException:
|
59 |
+
return 'Something unexpected happened, I\'m sorry. Try again.'
|
60 |
+
|
61 |
+
|
62 |
+
st.markdown('## Welcome to the CS article classification page!')
|
63 |
+
st.markdown('### What\'s below is pretty much self-explanatory.')
|
64 |
+
|
65 |
+
img_source = 'https://sun9-55.userapi.com/impg/azBQ_VTvbgEVonbL9hhFEpwyKAhjAtpVl4H2GQ/I4Vq0H6c3UM.jpg'
|
66 |
+
img_params = 'size=1200x900&quality=96&sign=f42419d9cdbf6fe55016fb002e4e85ae&type=album'
|
67 |
+
st.markdown(
|
68 |
+
f'<img src="{img_source}?{img_params}" width="70%"><br>',
|
69 |
+
unsafe_allow_html=True
|
70 |
+
)
|
71 |
+
|
72 |
+
title = st.text_input(
|
73 |
+
'Please, insert the title of the CS article you are interested in.',
|
74 |
+
placeholder='The title (e. g. Incorporating alien technologies in CV)'
|
75 |
+
)
|
76 |
+
|
77 |
+
summary = st.text_area(
|
78 |
+
'Now, please, insert the summary of the CS article you are interested in.',
|
79 |
+
height=250, placeholder='The summary itself.'
|
80 |
+
)
|
81 |
+
|
82 |
+
the_pipeline, cat_mapping, cat_name_mapping = initialize()
|
83 |
+
scores = get_top(the_pipeline, cat_mapping, title, summary)
|
84 |
+
|
85 |
+
if isinstance(scores, str):
|
86 |
+
st.markdown(scores)
|
87 |
+
else:
|
88 |
+
for score in scores:
|
89 |
+
percent = round(score['score'] * 100, 2)
|
90 |
+
category_short = score['label']
|
91 |
+
category_full = cat_name_mapping[category_short]
|
92 |
+
st.markdown(f'I\'m {percent}\% certain that the article is from the {category_short} category, which is "{category_full}"')
|
cat_mapping.json
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"0": "cs.AI",
|
3 |
+
"1": "cs.AR",
|
4 |
+
"2": "cs.CC",
|
5 |
+
"3": "cs.CE",
|
6 |
+
"4": "cs.CG",
|
7 |
+
"5": "cs.CL",
|
8 |
+
"6": "cs.CR",
|
9 |
+
"7": "cs.CV",
|
10 |
+
"8": "cs.CY",
|
11 |
+
"9": "cs.DB",
|
12 |
+
"10": "cs.DC",
|
13 |
+
"11": "cs.DL",
|
14 |
+
"12": "cs.DM",
|
15 |
+
"13": "cs.DS",
|
16 |
+
"14": "cs.ET",
|
17 |
+
"15": "cs.FL",
|
18 |
+
"16": "cs.GL",
|
19 |
+
"17": "cs.GR",
|
20 |
+
"18": "cs.GT",
|
21 |
+
"19": "cs.HC",
|
22 |
+
"20": "cs.IR",
|
23 |
+
"21": "cs.IT",
|
24 |
+
"22": "cs.LG",
|
25 |
+
"23": "cs.LO",
|
26 |
+
"24": "cs.MA",
|
27 |
+
"25": "cs.MM",
|
28 |
+
"26": "cs.MS",
|
29 |
+
"27": "cs.NA",
|
30 |
+
"28": "cs.NE",
|
31 |
+
"29": "cs.NI",
|
32 |
+
"30": "cs.OH",
|
33 |
+
"31": "cs.OS",
|
34 |
+
"32": "cs.PF",
|
35 |
+
"33": "cs.PL",
|
36 |
+
"34": "cs.RO",
|
37 |
+
"35": "cs.SC",
|
38 |
+
"36": "cs.SD",
|
39 |
+
"37": "cs.SE",
|
40 |
+
"38": "cs.SI",
|
41 |
+
"39": "cs.SY"
|
42 |
+
}
|
cat_name_mapping.json
ADDED
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"cs.AI": "Artificial Intelligence",
|
3 |
+
"cs.AR": "Hardware Architecture",
|
4 |
+
"cs.CC": "Computational Complexity",
|
5 |
+
"cs.CE": "Computational Engineering, Finance, and Science",
|
6 |
+
"cs.CG": "Computational Geometry",
|
7 |
+
"cs.CL": "Computation and Language",
|
8 |
+
"cs.CR": "Cryptography and Security",
|
9 |
+
"cs.CV": "Computer Vision and Pattern Recognition",
|
10 |
+
"cs.CY": "Computers and Society",
|
11 |
+
"cs.DB": "Databases",
|
12 |
+
"cs.DC": "Distributed, Parallel, and Cluster Computing",
|
13 |
+
"cs.DL": "Digital Libraries",
|
14 |
+
"cs.DM": "Discrete Mathematics",
|
15 |
+
"cs.DS": "Data Structures and Algorithms",
|
16 |
+
"cs.ET": "Emerging Technologies",
|
17 |
+
"cs.FL": "Formal Languages and Automata Theory",
|
18 |
+
"cs.GL": "General Literature",
|
19 |
+
"cs.GR": "Graphics",
|
20 |
+
"cs.GT": "Computer Science and Game Theory",
|
21 |
+
"cs.HC": "Human-Computer Interaction",
|
22 |
+
"cs.IR": "Information Retrieval",
|
23 |
+
"cs.IT": "Information Theory",
|
24 |
+
"cs.LG": "Machine Learning",
|
25 |
+
"cs.LO": "Logic in Computer Science",
|
26 |
+
"cs.MA": "Multiagent Systems",
|
27 |
+
"cs.MM": "Multimedia",
|
28 |
+
"cs.MS": "Mathematical Software",
|
29 |
+
"cs.NA": "Numerical Analysis",
|
30 |
+
"cs.NE": "Neural and Evolutionary Computing",
|
31 |
+
"cs.NI": "Networking and Internet Architecture",
|
32 |
+
"cs.OH": "Other Computer Science",
|
33 |
+
"cs.OS": "Operating Systems",
|
34 |
+
"cs.PF": "Performance",
|
35 |
+
"cs.PL": "Programming Languages",
|
36 |
+
"cs.RO": "Robotics",
|
37 |
+
"cs.SC": "Symbolic Computation",
|
38 |
+
"cs.SD": "Sound",
|
39 |
+
"cs.SE": "Software Engineering",
|
40 |
+
"cs.SI": "Social and Information Networks",
|
41 |
+
"cs.SY": "Systems and Control"
|
42 |
+
}
|
final_model/config.json
ADDED
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"activation": "gelu",
|
3 |
+
"architectures": [
|
4 |
+
"DistilBertForSequenceClassification"
|
5 |
+
],
|
6 |
+
"attention_dropout": 0.1,
|
7 |
+
"dim": 768,
|
8 |
+
"dropout": 0.1,
|
9 |
+
"hidden_dim": 3072,
|
10 |
+
"id2label": {
|
11 |
+
"0": "LABEL_0",
|
12 |
+
"1": "LABEL_1",
|
13 |
+
"2": "LABEL_2",
|
14 |
+
"3": "LABEL_3",
|
15 |
+
"4": "LABEL_4",
|
16 |
+
"5": "LABEL_5",
|
17 |
+
"6": "LABEL_6",
|
18 |
+
"7": "LABEL_7",
|
19 |
+
"8": "LABEL_8",
|
20 |
+
"9": "LABEL_9",
|
21 |
+
"10": "LABEL_10",
|
22 |
+
"11": "LABEL_11",
|
23 |
+
"12": "LABEL_12",
|
24 |
+
"13": "LABEL_13",
|
25 |
+
"14": "LABEL_14",
|
26 |
+
"15": "LABEL_15",
|
27 |
+
"16": "LABEL_16",
|
28 |
+
"17": "LABEL_17",
|
29 |
+
"18": "LABEL_18",
|
30 |
+
"19": "LABEL_19",
|
31 |
+
"20": "LABEL_20",
|
32 |
+
"21": "LABEL_21",
|
33 |
+
"22": "LABEL_22",
|
34 |
+
"23": "LABEL_23",
|
35 |
+
"24": "LABEL_24",
|
36 |
+
"25": "LABEL_25",
|
37 |
+
"26": "LABEL_26",
|
38 |
+
"27": "LABEL_27",
|
39 |
+
"28": "LABEL_28",
|
40 |
+
"29": "LABEL_29",
|
41 |
+
"30": "LABEL_30",
|
42 |
+
"31": "LABEL_31",
|
43 |
+
"32": "LABEL_32",
|
44 |
+
"33": "LABEL_33",
|
45 |
+
"34": "LABEL_34",
|
46 |
+
"35": "LABEL_35",
|
47 |
+
"36": "LABEL_36",
|
48 |
+
"37": "LABEL_37",
|
49 |
+
"38": "LABEL_38",
|
50 |
+
"39": "LABEL_39"
|
51 |
+
},
|
52 |
+
"initializer_range": 0.02,
|
53 |
+
"label2id": {
|
54 |
+
"LABEL_0": 0,
|
55 |
+
"LABEL_1": 1,
|
56 |
+
"LABEL_10": 10,
|
57 |
+
"LABEL_11": 11,
|
58 |
+
"LABEL_12": 12,
|
59 |
+
"LABEL_13": 13,
|
60 |
+
"LABEL_14": 14,
|
61 |
+
"LABEL_15": 15,
|
62 |
+
"LABEL_16": 16,
|
63 |
+
"LABEL_17": 17,
|
64 |
+
"LABEL_18": 18,
|
65 |
+
"LABEL_19": 19,
|
66 |
+
"LABEL_2": 2,
|
67 |
+
"LABEL_20": 20,
|
68 |
+
"LABEL_21": 21,
|
69 |
+
"LABEL_22": 22,
|
70 |
+
"LABEL_23": 23,
|
71 |
+
"LABEL_24": 24,
|
72 |
+
"LABEL_25": 25,
|
73 |
+
"LABEL_26": 26,
|
74 |
+
"LABEL_27": 27,
|
75 |
+
"LABEL_28": 28,
|
76 |
+
"LABEL_29": 29,
|
77 |
+
"LABEL_3": 3,
|
78 |
+
"LABEL_30": 30,
|
79 |
+
"LABEL_31": 31,
|
80 |
+
"LABEL_32": 32,
|
81 |
+
"LABEL_33": 33,
|
82 |
+
"LABEL_34": 34,
|
83 |
+
"LABEL_35": 35,
|
84 |
+
"LABEL_36": 36,
|
85 |
+
"LABEL_37": 37,
|
86 |
+
"LABEL_38": 38,
|
87 |
+
"LABEL_39": 39,
|
88 |
+
"LABEL_4": 4,
|
89 |
+
"LABEL_5": 5,
|
90 |
+
"LABEL_6": 6,
|
91 |
+
"LABEL_7": 7,
|
92 |
+
"LABEL_8": 8,
|
93 |
+
"LABEL_9": 9
|
94 |
+
},
|
95 |
+
"max_position_embeddings": 512,
|
96 |
+
"model_type": "distilbert",
|
97 |
+
"n_heads": 12,
|
98 |
+
"n_layers": 6,
|
99 |
+
"output_past": true,
|
100 |
+
"pad_token_id": 0,
|
101 |
+
"problem_type": "single_label_classification",
|
102 |
+
"qa_dropout": 0.1,
|
103 |
+
"seq_classif_dropout": 0.2,
|
104 |
+
"sinusoidal_pos_embds": false,
|
105 |
+
"tie_weights_": true,
|
106 |
+
"torch_dtype": "float32",
|
107 |
+
"transformers_version": "4.14.0",
|
108 |
+
"vocab_size": 28996
|
109 |
+
}
|
final_model/pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ba0c18f88b4a29acdd7ff9db7f997edd994d454382b0eda2c134b2b5a6022cff
|
3 |
+
size 263289073
|
requirements.txt
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
torch
|
2 |
+
transformers
|