Datasets:

Modalities:
Text
Formats:
parquet
Languages:
Spanish
Libraries:
Datasets
pandas
License:
RafaelJaime commited on
Commit
396b136
·
verified ·
1 Parent(s): a6e1a6e

Create generation_script.py

Browse files
Files changed (1) hide show
  1. generation_script.py +342 -0
generation_script.py ADDED
@@ -0,0 +1,342 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ !pip install PyPDF2 google google-genai requests python-dotenv datasets huggingface_hub
2
+
3
+ """# Libraries"""
4
+
5
+ import os
6
+ import requests
7
+ import json
8
+ from bs4 import BeautifulSoup
9
+ import PyPDF2
10
+ from google import genai
11
+ from google.genai import types
12
+ from dotenv import load_dotenv
13
+ import pandas as pd
14
+ from datasets import Dataset
15
+ from huggingface_hub import login
16
+
17
+ """# Get files"""
18
+
19
+ # Main page URL
20
+ main_url = 'https://www.sspa.juntadeandalucia.es/servicioandaluzdesalud/profesionales/ofertas-de-empleo/oferta-de-empleo-publico-puestos-base/oep-extraordinaria-decreto-ley-122022-centros-sas/cuadro-de-evolucion-concurso-oposicion-centros-sas'
21
+
22
+ # Main folder where exams will be saved
23
+ exams_folder = "exams"
24
+ os.makedirs(exams_folder, exist_ok=True)
25
+
26
+ # Perform an HTTP GET request to the main page
27
+ main_response = requests.get(main_url)
28
+
29
+ if main_response.status_code == 200:
30
+ main_soup = BeautifulSoup(main_response.content, 'html.parser')
31
+
32
+ # Find all tables on the main page
33
+ tables = main_soup.find_all('table')
34
+
35
+ for table in tables:
36
+ links = table.find_all('a', href=True)
37
+ for link in links:
38
+ secondary_url = link['href']
39
+ if secondary_url.startswith('/'):
40
+ secondary_url = 'https://www.sspa.juntadeandalucia.es' + secondary_url
41
+
42
+ folder_name = link.text.strip().replace("/", "-") # Replace invalid characters
43
+ folder_path = os.path.join(exams_folder, folder_name)
44
+ os.makedirs(folder_path, exist_ok=True)
45
+
46
+ secondary_response = requests.get(secondary_url)
47
+ if secondary_response.status_code == 200:
48
+ secondary_soup = BeautifulSoup(secondary_response.content, 'html.parser')
49
+ secondary_tables = secondary_soup.find_all('table')
50
+
51
+ for secondary_table in secondary_tables:
52
+ exam_booklet_links = secondary_table.find_all('a', title='Cuadernillo de Examen', href=True)
53
+ answer_sheet_links = secondary_table.find_all('a', title='Plantilla de respuestas', href=True)
54
+
55
+ for exam_booklet_link in exam_booklet_links:
56
+ pdf_url = exam_booklet_link['href']
57
+ if pdf_url.startswith('/'):
58
+ pdf_url = 'https://www.sspa.juntadeandalucia.es' + pdf_url
59
+ pdf_response = requests.get(pdf_url)
60
+ if pdf_response.status_code == 200:
61
+ file_path = os.path.join(folder_path, 'Exam_Booklet.pdf')
62
+ with open(file_path, 'wb') as pdf_file:
63
+ pdf_file.write(pdf_response.content)
64
+ print(f'Exam Booklet saved at: {file_path}')
65
+
66
+ for answer_sheet_link in answer_sheet_links:
67
+ pdf_url = answer_sheet_link['href']
68
+ if pdf_url.startswith('/'):
69
+ pdf_url = 'https://www.sspa.juntadeandalucia.es' + pdf_url
70
+ pdf_response = requests.get(pdf_url)
71
+ if pdf_response.status_code == 200:
72
+ file_path = os.path.join(folder_path, 'Answer_Sheet.pdf')
73
+ with open(file_path, 'wb') as pdf_file:
74
+ pdf_file.write(pdf_response.content)
75
+ print(f'Answer Sheet saved at: {file_path}')
76
+
77
+ else:
78
+ print(f'Error accessing the main page: {main_response.status_code}')
79
+
80
+ """# Configure apikey"""
81
+
82
+ load_dotenv()
83
+ GEMINI_API_KEY = "AIzaSyB46xwQbdXC_TK3Wnl3kt-oXtqKN8dRYfI"
84
+ # TODO: PONER AQUI EL TOKEN
85
+ HF_TOKEN= ""
86
+ HF_DATASET_NAME="RafaelJaime/sas_opposition_exam_data"
87
+
88
+ """# PDF processing
89
+
90
+ ## Extract text
91
+ """
92
+
93
+ def extract_text_from_pdf(pdf_path: str) -> str:
94
+ with open(pdf_path, "rb") as file:
95
+ reader = PyPDF2.PdfReader(file)
96
+ text = ""
97
+ for page in reader.pages:
98
+ text += page.extract_text()
99
+ return text
100
+
101
+ """## Number of questions"""
102
+
103
+ import base64
104
+ import os
105
+ from google import genai
106
+ from google.genai import types
107
+
108
+
109
+ def generate_number_questions(text):
110
+ client = genai.Client(
111
+ api_key=GEMINI_API_KEY,
112
+ )
113
+
114
+ model = "gemini-2.0-flash"
115
+ contents = [
116
+ types.Content(
117
+ role="user",
118
+ parts=[
119
+ types.Part.from_text(text=f"""tell me how many questions you have in format {{"number": "numberofquestionsinteger"}} in the following text: {text}"""),
120
+ ],
121
+ ),
122
+ ]
123
+ generate_content_config = types.GenerateContentConfig(
124
+ temperature=1,
125
+ top_p=0.95,
126
+ top_k=40,
127
+ max_output_tokens=8192,
128
+ response_mime_type="application/json",
129
+ )
130
+ response = client.models.generate_content(
131
+ model=model,
132
+ contents=contents,
133
+ config=generate_content_config,
134
+ )
135
+
136
+ return response.candidates[0].content.parts[0].text
137
+
138
+ """## Process with llm"""
139
+
140
+ import json
141
+ import os
142
+ from google import genai
143
+ from google.genai import types
144
+
145
+ def process_with_gemini(text: str, start: int, end: int):
146
+ client = genai.Client(
147
+ api_key=GEMINI_API_KEY
148
+ )
149
+
150
+ model = "gemini-2.0-flash"
151
+ contents = [
152
+ types.Content(
153
+ role="user",
154
+ parts=[
155
+ types.Part.from_text(text=f"""
156
+ Given the following text of an exam with questions and answers, extract each question and its possible answers.
157
+ Format the output as a list of JSON with the following format, I want you to extract questions from {start} to {end}:
158
+
159
+
160
+ {{{{"question number in integer format": {{"statement": "question text", "answers": ["option A", "option B", ...]}}}}}}
161
+
162
+ Exam text:
163
+ {text}
164
+ """),
165
+ ],
166
+ ),
167
+ ]
168
+
169
+ generate_content_config = types.GenerateContentConfig(
170
+ temperature=1,
171
+ top_p=0.95,
172
+ top_k=40,
173
+ max_output_tokens=32768,
174
+ response_mime_type="application/json",
175
+ )
176
+
177
+ # Use generate_content() instead of streaming
178
+ response = client.models.generate_content(
179
+ model=model,
180
+ contents=contents,
181
+ config=generate_content_config,
182
+ )
183
+
184
+ return response.text # Return the response instead of printing it
185
+
186
+ """# Collect the questions"""
187
+
188
+ import json
189
+ import os
190
+ from google import genai
191
+ from google.genai import types
192
+
193
+ def process_answers_with_gemini(text: str):
194
+ client = genai.Client(
195
+ api_key=GEMINI_API_KEY
196
+ )
197
+
198
+ model = "gemini-2.0-flash"
199
+ contents = [
200
+ types.Content(
201
+ role="user",
202
+ parts=[
203
+ types.Part.from_text(text=f"""
204
+ Please return the question number and the correct answers in format ['question number': 'answer letter','question number': 'answer letter'] from the following text
205
+ {text}
206
+ """),
207
+ ],
208
+ ),
209
+ ]
210
+
211
+ generate_content_config = types.GenerateContentConfig(
212
+ temperature=1,
213
+ top_p=0.95,
214
+ top_k=40,
215
+ max_output_tokens=32768,
216
+ response_mime_type="application/json",
217
+ )
218
+
219
+ # Use generate_content() instead of streaming
220
+ response = client.models.generate_content(
221
+ model=model,
222
+ contents=contents,
223
+ config=generate_content_config,
224
+ )
225
+
226
+ return response.text # Return the response instead of printing it
227
+
228
+ def process_pdf_file(pdf_path: str, answers_pdf_path: str, theme: str) -> pd.DataFrame:
229
+ pdf_text = extract_text_from_pdf(pdf_path)
230
+ result = generate_number_questions(pdf_text)
231
+ question_text = extract_text_from_pdf(answers_pdf_path)
232
+
233
+ # Process number of questions
234
+ try:
235
+ result_dict = json.loads(result)
236
+ except json.JSONDecodeError:
237
+ print("Error: The question count response is not valid JSON.")
238
+ return pd.DataFrame()
239
+
240
+ question_count = result_dict.get("number", "unknown")
241
+ print(f"The exam {pdf_path} contains {question_count} questions.")
242
+
243
+ try:
244
+ question_count = int(result_dict.get("number", 0))
245
+ except ValueError:
246
+ print(f"Error: Could not convert question count '{question_count}' to integer.")
247
+ return pd.DataFrame()
248
+
249
+ # Process questions in batches
250
+ questions = []
251
+ batch_size = 50
252
+ for start in range(1, question_count + 1, batch_size):
253
+ end = min(start + batch_size - 1, question_count)
254
+ print(f"Processing questions from {pdf_path} {start}-{end}...")
255
+ questions_subset = process_with_gemini(pdf_text, start, end)
256
+ questions.append(questions_subset)
257
+
258
+
259
+ # Combine all processed question batches
260
+ all_questions = []
261
+ for question_set in questions:
262
+ try:
263
+ question_list = json.loads(question_set)
264
+ all_questions.extend(question_list)
265
+ except json.JSONDecodeError:
266
+ print(f"Error: A question batch response is not valid JSON.")
267
+ continue
268
+
269
+ # If no valid questions were processed, return empty DataFrame
270
+ if not all_questions:
271
+ print("Error: No valid questions were processed.")
272
+ return pd.DataFrame()
273
+
274
+ # Process questions answers
275
+ questions_answer = process_answers_with_gemini(question_text)
276
+ try:
277
+ json_questions_answers = json.loads(questions_answer)
278
+ except json.JSONDecodeError:
279
+ print("Error: The response is not a valid JSON.")
280
+
281
+ # Format the data for the DataFrame
282
+ processed_data = []
283
+ for item in all_questions:
284
+ for key, value in item.items():
285
+ try:
286
+ real_answer = json_questions_answers[0].get(str(key), "Not available")
287
+ processed_data.append({
288
+ 'id': key,
289
+ 'statement': value['statement'],
290
+ 'answers': value['answers'],
291
+ 'real_answer': real_answer,
292
+ 'theme': theme
293
+ })
294
+ except KeyError as e:
295
+ print(f"Error: Missing key in question data: {e}")
296
+ # Skip this question but continue with others
297
+ continue
298
+
299
+ # Create DataFrame from dictionary list
300
+ df = pd.DataFrame(processed_data)
301
+ if not df.empty:
302
+ df.set_index('id', inplace=True)
303
+
304
+ return df
305
+
306
+
307
+ all_df_array = []
308
+ # Verify that the folder exists
309
+ if os.path.exists(exams_folder):
310
+ for folder_name in os.listdir(exams_folder):
311
+ folder_path = os.path.join(exams_folder, folder_name)
312
+
313
+ # Verify that it's a folder
314
+ if os.path.isdir(folder_path):
315
+ print(f"Processing: {folder_name}")
316
+
317
+ files = os.listdir(folder_path)
318
+
319
+ # Initialize question and answer paths
320
+ questions_path = None
321
+ answers_path = None
322
+
323
+ # Look for files that start with the desired prefixes
324
+ for file in files:
325
+ if file.startswith('Exam_Booklet') and not questions_path:
326
+ questions_path = os.path.join(folder_path, file)
327
+ elif file.startswith('Answer_Sheet') and not answers_path:
328
+ answers_path = os.path.join(folder_path, file)
329
+ exam_df = process_pdf_file(questions_path, answers_path, folder_name)
330
+ if not exam_df.empty:
331
+ all_df_array.append(exam_df)
332
+
333
+ if all_df_array:
334
+ df = pd.concat(all_df_array, ignore_index=True) # `ignore_index=True` to avoid duplicates in the index
335
+ print(f"Final DataFrame with all questions and answers:\n{df}")
336
+ else:
337
+ print("No valid DataFrames were generated.")
338
+
339
+ """# Upload to huggingface"""
340
+
341
+ login(HF_TOKEN)
342
+ Dataset.from_pandas(df).push_to_hub(HF_DATASET_NAME)