Spaces:
Sleeping
Sleeping
arithescientist
commited on
Commit
•
342a4a2
1
Parent(s):
02a288f
Update app.py
Browse files
app.py
CHANGED
@@ -17,275 +17,118 @@ import yake
|
|
17 |
from transformers import AutoTokenizer, AutoModelForPreTraining, AutoModel, AutoConfig
|
18 |
from summarizer import Summarizer,TransformerSummarizer
|
19 |
from transformers import pipelines
|
20 |
-
|
21 |
|
22 |
print("lets go")
|
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 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
# main index page route
|
65 |
-
@app.route('/')
|
66 |
-
@cross_origin()
|
67 |
-
def index():
|
68 |
-
return render_template('index.html')
|
69 |
-
|
70 |
-
@cross_origin()
|
71 |
-
@app.route('/results')
|
72 |
-
def results():
|
73 |
-
return render_template('results.html')
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
@app.route('/predict', methods=['GET', 'POST'])
|
78 |
-
def uploads():
|
79 |
-
if request.method == 'GET':
|
80 |
-
# Get the file from post request
|
81 |
-
|
82 |
-
numsent = int(request.args['number'])
|
83 |
-
text = str(request.args['text'])
|
84 |
-
content = text
|
85 |
-
|
86 |
-
|
87 |
-
summary_text = ""
|
88 |
-
for i, paragraph in enumerate(content.split("\n\n")):
|
89 |
-
|
90 |
-
paragraph = paragraph.replace('\n',' ')
|
91 |
-
paragraph = paragraph.replace('\t','')
|
92 |
-
paragraph = ' '.join(paragraph.split())
|
93 |
-
# count words in the paragraph and exclude if less than 4 words
|
94 |
-
tokens = word_tokenize(paragraph)
|
95 |
-
# only do real words
|
96 |
-
tokens = [word for word in tokens if word.isalpha()]
|
97 |
-
# print("\nTokens: {}\n".format(len(tokens)))
|
98 |
-
# only do sentences with more than 1 words excl. alpha crap
|
99 |
-
if len(tokens) <= 1:
|
100 |
-
continue
|
101 |
-
# Perhaps also ignore paragraphs with no sentence?
|
102 |
-
sentences = sent_tokenize(paragraph)
|
103 |
-
|
104 |
-
paragraph = ' '.join(tokens)
|
105 |
-
|
106 |
-
print("\nParagraph:")
|
107 |
-
print(paragraph+"\n")
|
108 |
-
# T5 needs to have 'summarize' in order to work:
|
109 |
-
# text = "summarize:" + paragraph
|
110 |
-
text = paragraph
|
111 |
-
|
112 |
-
summary = bert_legal_model(text, min_length = 8, ratio = 0.05)
|
113 |
-
# summary = tokenizer_t5.decode(summary_ids[0], skip_special_tokens=True)
|
114 |
-
summary_text += str(summary) + "\n\n"
|
115 |
-
print("Summary:")
|
116 |
-
print(summary)
|
117 |
-
|
118 |
-
content2 = content.replace('\n',' ')
|
119 |
-
content2 = content2.replace('\t','')
|
120 |
-
summary = bert_legal_model(content2, min_length = 8, num_sentences=25)
|
121 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
|
|
|
123 |
|
124 |
-
# write all to file for inspection and storage
|
125 |
-
all_text = "The Summary-- " + str(summary) + "\n\n\n" \
|
126 |
-
+ "The Larger Summary-- " + str(summary_text)
|
127 |
-
|
128 |
-
|
129 |
-
all_text2 = all_text.encode('latin-1', 'replace').decode('latin-1')
|
130 |
-
all_text2 = all_text2.replace('?','.')
|
131 |
-
all_text2 = all_text2.replace('\n',' ')
|
132 |
-
all_text2 = all_text2.replace('..','.')
|
133 |
-
all_text2 = all_text2.replace(',.',',')
|
134 |
-
all_text2 = all_text2.replace('-- ','\n\n\n')
|
135 |
-
|
136 |
-
pdf = FPDF()
|
137 |
-
|
138 |
-
# Add a page
|
139 |
-
pdf.add_page()
|
140 |
-
|
141 |
-
pdf.set_font("Times", size = 12)
|
142 |
-
|
143 |
-
# open the text file in read mode
|
144 |
-
f = all_text2
|
145 |
-
|
146 |
-
# insert the texts in pdf
|
147 |
-
pdf.multi_cell(190, 10, txt = f, align = 'C')
|
148 |
-
|
149 |
-
|
150 |
-
# save the pdf with name .pdf
|
151 |
-
pdf.output("./static/legal.pdf")
|
152 |
-
all_text
|
153 |
-
|
154 |
-
|
155 |
-
return render_template('results.html')
|
156 |
-
return None
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
@app.route('/predictpdf', methods=['GET', 'POST'])
|
162 |
-
def uploads2():
|
163 |
-
if request.method == 'POST':
|
164 |
-
# Get the file from post request
|
165 |
-
|
166 |
-
numsent = int(request.args['number'])
|
167 |
-
if 'file' not in request.files:
|
168 |
-
flash('No file part')
|
169 |
-
return redirect(request.url)
|
170 |
-
file = request.files['file']
|
171 |
-
# if user does not select file, browser also
|
172 |
-
# submit an empty part without filename
|
173 |
-
if file.filename == '':
|
174 |
-
flash('No selected file')
|
175 |
-
return redirect(request.url)
|
176 |
-
if file and allowed_file(file.filename):
|
177 |
-
filename = "legal.pdf"
|
178 |
-
file.save(os.path.join(app.config['UPLOAD_FOLDER'], filename))
|
179 |
-
|
180 |
-
f = request.files['file']
|
181 |
-
f.save(secure_filename(f.filename))
|
182 |
-
|
183 |
-
|
184 |
-
path = os.getcwd()
|
185 |
-
folder_name = 'pdfs'
|
186 |
-
path = os.path.join(path, folder_name)
|
187 |
-
|
188 |
-
list_of_files = []
|
189 |
-
for root, dirs, files in os.walk(path):
|
190 |
-
for file in files:
|
191 |
-
if(file.endswith(".pdf")):
|
192 |
-
# print(os.path.join(root,file))
|
193 |
-
list_of_files.append(os.path.join(root,file))
|
194 |
-
|
195 |
-
print("\nProcessing {} files...\n".format(len(list_of_files)))
|
196 |
-
total_pages = 0
|
197 |
-
|
198 |
-
for filename in list_of_files:
|
199 |
-
print(filename)
|
200 |
-
file = os.path.splitext(os.path.basename(filename))[0]
|
201 |
-
pages = pdf2image.convert_from_path(pdf_path=filename, dpi=400, size=(1654,2340))
|
202 |
-
total_pages += len(pages)
|
203 |
-
print("\nProcessing the next {} pages...\n".format(len(pages)))
|
204 |
-
|
205 |
-
# Then save all pages as images and convert them to text except the last page
|
206 |
-
# TODO: create this as a function
|
207 |
-
content = ""
|
208 |
-
dir_name = 'images/' + file + '/'
|
209 |
-
os.makedirs(dir_name, exist_ok=True)
|
210 |
-
# If folder doesn't exist, then create it.
|
211 |
-
for i in range(len(pages)-1):
|
212 |
-
pages[i].save(dir_name + str(i) + '.jpg')
|
213 |
-
# OCR the image using Google's tesseract
|
214 |
-
content += pt.image_to_string(pages[i])
|
215 |
-
|
216 |
-
summary_text = ""
|
217 |
-
for i, paragraph in enumerate(content.split("\n\n")):
|
218 |
-
|
219 |
-
paragraph = paragraph.replace('\n',' ')
|
220 |
-
paragraph = paragraph.replace('\t','')
|
221 |
-
paragraph = ' '.join(paragraph.split())
|
222 |
-
# count words in the paragraph and exclude if less than 4 words
|
223 |
-
tokens = word_tokenize(paragraph)
|
224 |
-
# only do real words
|
225 |
-
tokens = [word for word in tokens if word.isalpha()]
|
226 |
-
# print("\nTokens: {}\n".format(len(tokens)))
|
227 |
-
# only do sentences with more than 1 words excl. alpha crap
|
228 |
-
if len(tokens) <= 1:
|
229 |
-
continue
|
230 |
-
# Perhaps also ignore paragraphs with no sentence?
|
231 |
-
sentences = sent_tokenize(paragraph)
|
232 |
-
|
233 |
-
paragraph = ' '.join(tokens)
|
234 |
-
|
235 |
-
print("\nParagraph:")
|
236 |
-
print(paragraph+"\n")
|
237 |
-
# T5 needs to have 'summarize' in order to work:
|
238 |
-
# text = "summarize:" + paragraph
|
239 |
-
text = paragraph
|
240 |
-
|
241 |
-
summary = bert_legal_model(text, min_length = 8, ratio = 0.05)
|
242 |
-
# summary = tokenizer_t5.decode(summary_ids[0], skip_special_tokens=True)
|
243 |
-
summary_text += str(summary) + "\n\n"
|
244 |
-
print("Summary:")
|
245 |
-
print(summary)
|
246 |
-
|
247 |
-
content2 = content.replace('\n',' ')
|
248 |
-
content2 = content2.replace('\t','')
|
249 |
-
summary = bert_legal_model(content2, min_length = 8, num_sentences=25)
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
# write all to file for inspection and storage
|
254 |
-
all_text = "The Summary-- " + str(summary) + "\n\n\n" \
|
255 |
-
+ "The Larger Summary-- " + str(summary_text)
|
256 |
-
|
257 |
-
|
258 |
-
all_text2 = all_text.encode('latin-1', 'replace').decode('latin-1')
|
259 |
-
all_text2 = all_text2.replace('?','.')
|
260 |
-
all_text2 = all_text2.replace('\n',' ')
|
261 |
-
all_text2 = all_text2.replace('..','.')
|
262 |
-
all_text2 = all_text2.replace(',.',',')
|
263 |
-
all_text2 = all_text2.replace('-- ','\n\n\n')
|
264 |
-
|
265 |
-
pdf = FPDF()
|
266 |
-
|
267 |
-
# Add a page
|
268 |
-
pdf.add_page()
|
269 |
-
|
270 |
-
pdf.set_font("Times", size = 12)
|
271 |
-
|
272 |
-
# open the text file in read mode
|
273 |
-
f = all_text2
|
274 |
-
|
275 |
-
# insert the texts in pdf
|
276 |
-
pdf.multi_cell(190, 10, txt = f, align = 'C')
|
277 |
-
|
278 |
-
|
279 |
-
# save the pdf with name .pdf
|
280 |
-
pdf.output("./static/legal.pdf")
|
281 |
-
all_text
|
282 |
-
|
283 |
-
|
284 |
-
return render_template('results.html')
|
285 |
-
return None
|
286 |
|
287 |
|
288 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
289 |
|
290 |
-
|
291 |
-
iface.launch()
|
|
|
17 |
from transformers import AutoTokenizer, AutoModelForPreTraining, AutoModel, AutoConfig
|
18 |
from summarizer import Summarizer,TransformerSummarizer
|
19 |
from transformers import pipelines
|
20 |
+
nltk.download('punkt')
|
21 |
|
22 |
print("lets go")
|
23 |
|
24 |
+
def pdf(file):
|
25 |
+
#model_name = 'laxya007/gpt2_legal'
|
26 |
+
# model_name = 'facebook/bart-large-cnn'
|
27 |
+
model_name = 'nlpaueb/legal-bert-base-uncased'
|
28 |
+
|
29 |
+
# The setup of huggingface.co
|
30 |
+
custom_config = AutoConfig.from_pretrained(model_name)
|
31 |
+
custom_config.output_hidden_states=True
|
32 |
+
custom_tokenizer = AutoTokenizer.from_pretrained(model_name)
|
33 |
+
custom_model = AutoModel.from_pretrained(model_name, config=custom_config)
|
34 |
+
bert_legal_model = Summarizer(custom_model=custom_model, custom_tokenizer=custom_tokenizer)
|
35 |
+
print('Using model {}\n'.format(model_name))
|
36 |
+
|
37 |
+
list_of_files = file
|
38 |
+
|
39 |
+
|
40 |
+
print("\nProcessing {} files...\n".format(len(list_of_files)))
|
41 |
+
total_pages = 0
|
42 |
+
|
43 |
+
for filename in list_of_files:
|
44 |
+
print(filename)
|
45 |
+
file = os.path.splitext(os.path.basename(filename))[0]
|
46 |
+
pages = pdf2image.convert_from_path(pdf_path=filename, dpi=400, size=(1654,2340))
|
47 |
+
total_pages += len(pages)
|
48 |
+
print("\nProcessing the next {} pages...\n".format(len(pages)))
|
49 |
+
|
50 |
+
# Then save all pages as images and convert them to text except the last page
|
51 |
+
# TODO: create this as a function
|
52 |
+
content = ""
|
53 |
+
dir_name = 'images/' + file + '/'
|
54 |
+
os.makedirs(dir_name, exist_ok=True)
|
55 |
+
# If folder doesn't exist, then create it.
|
56 |
+
for i in range(len(pages)-1):
|
57 |
+
pages[i].save(dir_name + str(i) + '.jpg')
|
58 |
+
# OCR the image using Google's tesseract
|
59 |
+
content += pt.image_to_string(pages[i])
|
60 |
+
|
61 |
+
summary_text = ""
|
62 |
+
for i, paragraph in enumerate(content.split("\n\n")):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
|
64 |
+
paragraph = paragraph.replace('\n',' ')
|
65 |
+
paragraph = paragraph.replace('\t','')
|
66 |
+
paragraph = ' '.join(paragraph.split())
|
67 |
+
# count words in the paragraph and exclude if less than 4 words
|
68 |
+
tokens = word_tokenize(paragraph)
|
69 |
+
# only do real words
|
70 |
+
tokens = [word for word in tokens if word.isalpha()]
|
71 |
+
# print("\nTokens: {}\n".format(len(tokens)))
|
72 |
+
# only do sentences with more than 1 words excl. alpha crap
|
73 |
+
if len(tokens) <= 1:
|
74 |
+
continue
|
75 |
+
# Perhaps also ignore paragraphs with no sentence?
|
76 |
+
sentences = sent_tokenize(paragraph)
|
77 |
+
|
78 |
+
paragraph = ' '.join(tokens)
|
79 |
+
|
80 |
+
print("\nParagraph:")
|
81 |
+
print(paragraph+"\n")
|
82 |
+
# T5 needs to have 'summarize' in order to work:
|
83 |
+
# text = "summarize:" + paragraph
|
84 |
+
text = paragraph
|
85 |
+
|
86 |
+
summary = bert_legal_model(text, min_length = 8, ratio = 0.05)
|
87 |
+
# summary = tokenizer_t5.decode(summary_ids[0], skip_special_tokens=True)
|
88 |
+
summary_text += str(summary) + "\n\n"
|
89 |
+
print("Summary:")
|
90 |
+
print(summary)
|
91 |
+
|
92 |
+
content2 = content.replace('\n',' ')
|
93 |
+
content2 = content2.replace('\t','')
|
94 |
+
summary = bert_legal_model(content2, min_length = 8, num_sentences=25)
|
95 |
+
|
96 |
+
|
97 |
+
|
98 |
+
# write all to file for inspection and storage
|
99 |
+
all_text = "The Summary-- " + str(summary) + "\n\n\n" \
|
100 |
+
+ "The Larger Summary-- " + str(summary_text)
|
101 |
+
|
102 |
+
|
103 |
+
all_text2 = all_text.encode('latin-1', 'replace').decode('latin-1')
|
104 |
+
all_text2 = all_text2.replace('?','.')
|
105 |
+
all_text2 = all_text2.replace('\n',' ')
|
106 |
+
all_text2 = all_text2.replace('..','.')
|
107 |
+
all_text2 = all_text2.replace(',.',',')
|
108 |
+
all_text2 = all_text2.replace('-- ','\n\n\n')
|
109 |
+
|
110 |
+
pdf = FPDF()
|
111 |
+
|
112 |
+
# Add a page
|
113 |
+
pdf.add_page()
|
114 |
+
|
115 |
+
pdf.set_font("Times", size = 12)
|
116 |
+
|
117 |
+
# open the text file in read mode
|
118 |
+
f = all_text2
|
119 |
+
return f
|
120 |
|
121 |
+
import gradio as gr
|
122 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
123 |
|
124 |
|
125 |
+
iface = gr.Interface(
|
126 |
+
pdf,
|
127 |
+
gr.inputs.Image(shape=(224, 224)),
|
128 |
+
gr.outputs.Label(f),
|
129 |
+
capture_session=True,
|
130 |
+
interpretation="default",
|
131 |
+
)
|
132 |
|
133 |
+
if __name__ == "__main__":
|
134 |
+
iface.launch(share=True)
|