Datasets:
Tasks:
Text Classification
Modalities:
Text
Formats:
parquet
Languages:
Portuguese
Size:
1K - 10K
License:
Delete loading script
Browse files- aes_enem_dataset.py +0 -1234
aes_enem_dataset.py
DELETED
@@ -1,1234 +0,0 @@
|
|
1 |
-
# Copyright 2023 Andre Barbosa, Igor Cataneo Silveira & The HuggingFace Datasets Authors
|
2 |
-
#
|
3 |
-
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
-
# you may not use this file except in compliance with the License.
|
5 |
-
# You may obtain a copy of the License at
|
6 |
-
#
|
7 |
-
# http://www.apache.org/licenses/LICENSE-2.0
|
8 |
-
#
|
9 |
-
# Unless required by applicable law or agreed to in writing, software
|
10 |
-
# distributed under the License is distributed on an "AS IS" BASIS,
|
11 |
-
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
-
# See the License for the specific language governing permissions and
|
13 |
-
# limitations under the License.
|
14 |
-
|
15 |
-
import csv
|
16 |
-
import math
|
17 |
-
import os
|
18 |
-
import re
|
19 |
-
from pathlib import Path
|
20 |
-
|
21 |
-
import datasets
|
22 |
-
import numpy as np
|
23 |
-
import pandas as pd
|
24 |
-
from multiprocessing import Pool, cpu_count
|
25 |
-
from bs4 import BeautifulSoup
|
26 |
-
from tqdm.auto import tqdm
|
27 |
-
|
28 |
-
RANDOM_STATE = 42
|
29 |
-
np.random.seed(RANDOM_STATE) # Set the seed
|
30 |
-
|
31 |
-
_CITATION = """
|
32 |
-
@inproceedings{silveira-etal-2024-new,
|
33 |
-
title = "A New Benchmark for Automatic Essay Scoring in {P}ortuguese",
|
34 |
-
author = "Silveira, Igor Cataneo and
|
35 |
-
Barbosa, Andr{\'e} and
|
36 |
-
Mau{\'a}, Denis Deratani",
|
37 |
-
editor = "Gamallo, Pablo and
|
38 |
-
Claro, Daniela and
|
39 |
-
Teixeira, Ant{\'o}nio and
|
40 |
-
Real, Livy and
|
41 |
-
Garcia, Marcos and
|
42 |
-
Oliveira, Hugo Goncalo and
|
43 |
-
Amaro, Raquel",
|
44 |
-
booktitle = "Proceedings of the 16th International Conference on Computational Processing of Portuguese - Vol. 1",
|
45 |
-
month = mar,
|
46 |
-
year = "2024",
|
47 |
-
address = "Santiago de Compostela, Galicia/Spain",
|
48 |
-
publisher = "Association for Computational Lingustics",
|
49 |
-
url = "https://aclanthology.org/2024.propor-1.23/",
|
50 |
-
pages = "228--237"
|
51 |
-
}
|
52 |
-
"""
|
53 |
-
|
54 |
-
_DESCRIPTION = """\
|
55 |
-
This dataset was created as part of our work on advancing Automatic Essay Scoring for
|
56 |
-
Brazilian Portuguese. It comprises a large collection of publicly available essays
|
57 |
-
collected from websites simulating University Entrance Exams, with a subset expertly
|
58 |
-
annotated to provide reliable assessment indicators. The dataset includes both the raw
|
59 |
-
text and processed forms of the essays, along with supporting prompts and supplemental
|
60 |
-
texts.
|
61 |
-
|
62 |
-
Key Features:
|
63 |
-
- A diverse corpus of essays with detailed annotations.
|
64 |
-
- A subset graded by expert annotators to evaluate essay quality and task difficulty.
|
65 |
-
- Comprehensive metadata providing provenance and context for each essay.
|
66 |
-
- An empirical analysis framework to support state-of-the-art predictive modeling.
|
67 |
-
|
68 |
-
For further details, please refer to the paper “A New Benchmark for Automatic Essay
|
69 |
-
Scoring in Portuguese” available at https://aclanthology.org/2024.propor-1.23/.
|
70 |
-
"""
|
71 |
-
|
72 |
-
# TODO: Add a link to an official homepage for the dataset here
|
73 |
-
_HOMEPAGE = ""
|
74 |
-
|
75 |
-
# TODO: Add the licence for the dataset here if you can find it
|
76 |
-
_LICENSE = ""
|
77 |
-
|
78 |
-
|
79 |
-
_URLS = {
|
80 |
-
"sourceAOnly": "https://huggingface.co/datasets/kamel-usp/aes_enem_dataset/resolve/main/sourceAWithGraders.tar.gz",
|
81 |
-
"sourceAWithGraders": "https://huggingface.co/datasets/kamel-usp/aes_enem_dataset/resolve/main/sourceAWithGraders.tar.gz",
|
82 |
-
"sourceB": "https://huggingface.co/datasets/kamel-usp/aes_enem_dataset/resolve/main/sourceB.tar.gz",
|
83 |
-
"PROPOR2024": "https://huggingface.co/datasets/kamel-usp/aes_enem_dataset/resolve/main/propor2024.tar.gz",
|
84 |
-
"gradesThousand": "https://huggingface.co/datasets/kamel-usp/aes_enem_dataset/resolve/main/scrapedGradesThousand.tar.gz",
|
85 |
-
}
|
86 |
-
|
87 |
-
|
88 |
-
PROMPTS_TO_IGNORE = [
|
89 |
-
"brasileiros-tem-pessima-educacao-argumentativa-segundo-cientista",
|
90 |
-
"carta-convite-discutir-discriminacao-na-escola",
|
91 |
-
"informacao-no-rotulo-de-produtos-transgenicos",
|
92 |
-
]
|
93 |
-
|
94 |
-
# Essays to Ignore
|
95 |
-
ESSAY_TO_IGNORE = [
|
96 |
-
"direitos-em-conflito-liberdade-de-expressao-e-intimidade/2.html",
|
97 |
-
"terceirizacao-avanco-ou-retrocesso/2.html",
|
98 |
-
"artes-e-educacao-fisica-opcionais-ou-obrigatorias/2.html",
|
99 |
-
"violencia-e-drogas-o-papel-do-usuario/0.html",
|
100 |
-
"internacao-compulsoria-de-dependentes-de-crack/0.html",
|
101 |
-
]
|
102 |
-
|
103 |
-
CSV_HEADER = [
|
104 |
-
"id",
|
105 |
-
"id_prompt",
|
106 |
-
"prompt",
|
107 |
-
"supporting_text",
|
108 |
-
"title",
|
109 |
-
"essay",
|
110 |
-
"grades",
|
111 |
-
"general",
|
112 |
-
"specific",
|
113 |
-
"essay_year",
|
114 |
-
"reference",
|
115 |
-
]
|
116 |
-
|
117 |
-
CSV_HEADERPROPOR = [
|
118 |
-
"id",
|
119 |
-
"id_prompt",
|
120 |
-
"title",
|
121 |
-
"essay",
|
122 |
-
"grades",
|
123 |
-
"essay_year",
|
124 |
-
"reference",
|
125 |
-
]
|
126 |
-
|
127 |
-
CSV_HEADERTHOUSAND = [
|
128 |
-
"id",
|
129 |
-
"author",
|
130 |
-
"id_prompt",
|
131 |
-
"essay_year",
|
132 |
-
"grades",
|
133 |
-
"essay",
|
134 |
-
"source",
|
135 |
-
"supporting_text",
|
136 |
-
"prompt",
|
137 |
-
]
|
138 |
-
|
139 |
-
CSV_HEADER_JBCS25 = [
|
140 |
-
"id",
|
141 |
-
"id_prompt",
|
142 |
-
"essay_text",
|
143 |
-
"grades",
|
144 |
-
"essay_year",
|
145 |
-
"supporting_text",
|
146 |
-
"prompt",
|
147 |
-
"reference",
|
148 |
-
]
|
149 |
-
|
150 |
-
SOURCE_A_DESC = """
|
151 |
-
SourceA have 860 essays available from August 2015 to March 2020.
|
152 |
-
For each month of that period, a new prompt together with supporting texts were given,
|
153 |
-
and the graded essays from the previous month were made available.
|
154 |
-
Of the 56 prompts, 12 had no associated essays available (at the time of download).
|
155 |
-
Additionally, there were 3 prompts that asked for a text in the format of a letter.
|
156 |
-
We removed those 15 prompts and associated texts from the corpus.
|
157 |
-
For an unknown reason, 414 of the essays were graded using a five-point scale of either
|
158 |
-
{0, 50, 100, 150, 200} or its scaled-down version going from 0 to 2.
|
159 |
-
To avoid introducing bias, we also discarded such instances, resulting in a dataset of
|
160 |
-
386 annotated essays with prompts and supporting texts (with each component being clearly identified).
|
161 |
-
Some of the essays used a six-point scale with 20 points instead of 40 points as the second class.
|
162 |
-
As we believe this introduces minimal bias, we kept such essays and relabeled class 20 as class 40.
|
163 |
-
The original data contains comments from the annotators explaining their per-competence scores.
|
164 |
-
They are included in our dataset.
|
165 |
-
"""
|
166 |
-
|
167 |
-
SOURCE_A_WITH_GRADERS = """
|
168 |
-
sourceAWithGraders includes the original dataset augmented with grades from additional reviewers.
|
169 |
-
Each essay is replicated three times:
|
170 |
-
1. The original essay with its grades from the website.
|
171 |
-
2. The same essay with grades from the first human grader.
|
172 |
-
3. The same essay with grades from the second human grader.
|
173 |
-
"""
|
174 |
-
|
175 |
-
SOURCE_B_DESC = """
|
176 |
-
SourceB is very similar to Source A: a new prompt and supporting texts are made
|
177 |
-
available every month along with the graded essays submitted in the previous month.
|
178 |
-
We downloaded HTML sources from 7,700 essays from May 2009 to May 2023. Essays released
|
179 |
-
prior to June 2016 were graded on a five-point scale and consequently discarded.
|
180 |
-
This resulted in a corpus of approx. 3,200 graded essays on 83 different prompts.
|
181 |
-
|
182 |
-
Although in principle, Source B also provides supporting texts for students, none were
|
183 |
-
available at the time the data was downloaded.
|
184 |
-
To mitigate this, we extracted supporting texts from the Essay-Br corpus, whenever
|
185 |
-
possible, by manually matching prompts between the two corpora.
|
186 |
-
We ended up with approx. 1,000 essays containing both prompt and supporting texts, and
|
187 |
-
approx. 2,200 essays containing only the respective prompt.
|
188 |
-
"""
|
189 |
-
|
190 |
-
PROPOR2024 = """
|
191 |
-
This split corresponds to the results reported in the PROPOR 2024 paper. While reproducibility was
|
192 |
-
fixed in the sourceAWithGraders configuration, this split preserves the original
|
193 |
-
distribution of prompts and scores as used in the paper.
|
194 |
-
"""
|
195 |
-
|
196 |
-
GRADES_THOUSAND = """
|
197 |
-
TODO
|
198 |
-
"""
|
199 |
-
|
200 |
-
JBCS2025 = """
|
201 |
-
TODO
|
202 |
-
"""
|
203 |
-
|
204 |
-
|
205 |
-
class AesEnemDataset(datasets.GeneratorBasedBuilder):
|
206 |
-
"""
|
207 |
-
AES Enem Dataset. For full explanation about generation process, please refer to: https://aclanthology.org/2024.propor-1.23/
|
208 |
-
|
209 |
-
We realized in our experiments that there was an issue in the determistic process regarding how the dataset is generated.
|
210 |
-
To reproduce results from PROPOR paper, please refer to "PROPOR2024" config. Other configs are reproducible now.
|
211 |
-
"""
|
212 |
-
|
213 |
-
VERSION = datasets.Version("1.0.0")
|
214 |
-
|
215 |
-
# You will be able to load one or the other configurations in the following list with
|
216 |
-
BUILDER_CONFIGS = [
|
217 |
-
datasets.BuilderConfig(
|
218 |
-
name="sourceAOnly", version=VERSION, description=SOURCE_A_DESC
|
219 |
-
),
|
220 |
-
datasets.BuilderConfig(
|
221 |
-
name="sourceAWithGraders",
|
222 |
-
version=VERSION,
|
223 |
-
description=SOURCE_A_WITH_GRADERS,
|
224 |
-
),
|
225 |
-
datasets.BuilderConfig(
|
226 |
-
name="sourceB",
|
227 |
-
version=VERSION,
|
228 |
-
description=SOURCE_B_DESC,
|
229 |
-
),
|
230 |
-
datasets.BuilderConfig(
|
231 |
-
name="PROPOR2024", version=VERSION, description=PROPOR2024
|
232 |
-
),
|
233 |
-
datasets.BuilderConfig(
|
234 |
-
name="gradesThousand", version=VERSION, description=GRADES_THOUSAND
|
235 |
-
),
|
236 |
-
datasets.BuilderConfig(name="JBCS2025", version=VERSION, description=JBCS2025),
|
237 |
-
]
|
238 |
-
|
239 |
-
def _info(self):
|
240 |
-
if self.config.name == "PROPOR2024":
|
241 |
-
features = datasets.Features(
|
242 |
-
{
|
243 |
-
"id": datasets.Value("string"),
|
244 |
-
"id_prompt": datasets.Value("string"),
|
245 |
-
"essay_title": datasets.Value("string"),
|
246 |
-
"essay_text": datasets.Value("string"),
|
247 |
-
"grades": datasets.Sequence(datasets.Value("int16")),
|
248 |
-
"essay_year": datasets.Value("int16"),
|
249 |
-
"reference": datasets.Value("string"),
|
250 |
-
}
|
251 |
-
)
|
252 |
-
elif self.config.name == "gradesThousand":
|
253 |
-
features = datasets.Features(
|
254 |
-
{
|
255 |
-
"id": datasets.Value("string"),
|
256 |
-
"id_prompt": datasets.Value("string"),
|
257 |
-
"supporting_text": datasets.Value("string"),
|
258 |
-
"prompt": datasets.Value("string"),
|
259 |
-
"essay_text": datasets.Value("string"),
|
260 |
-
"grades": datasets.Sequence(datasets.Value("int16")),
|
261 |
-
"essay_year": datasets.Value("int16"),
|
262 |
-
"source": datasets.Value("string"),
|
263 |
-
}
|
264 |
-
)
|
265 |
-
elif self.config.name == "JBCS2025":
|
266 |
-
features = datasets.Features(
|
267 |
-
{
|
268 |
-
"id": datasets.Value("string"),
|
269 |
-
"id_prompt": datasets.Value("string"),
|
270 |
-
"essay_text": datasets.Value("string"),
|
271 |
-
"grades": datasets.Sequence(datasets.Value("int16")),
|
272 |
-
"essay_year": datasets.Value("int16"),
|
273 |
-
"supporting_text": datasets.Value("string"),
|
274 |
-
"prompt": datasets.Value("string"),
|
275 |
-
"reference": datasets.Value("string"),
|
276 |
-
}
|
277 |
-
)
|
278 |
-
else:
|
279 |
-
features = datasets.Features(
|
280 |
-
{
|
281 |
-
"id": datasets.Value("string"),
|
282 |
-
"id_prompt": datasets.Value("string"),
|
283 |
-
"prompt": datasets.Value("string"),
|
284 |
-
"supporting_text": datasets.Value("string"),
|
285 |
-
"essay_title": datasets.Value("string"),
|
286 |
-
"essay_text": datasets.Value("string"),
|
287 |
-
"grades": datasets.Sequence(datasets.Value("int16")),
|
288 |
-
"essay_year": datasets.Value("int16"),
|
289 |
-
"general_comment": datasets.Value("string"),
|
290 |
-
"specific_comment": datasets.Value("string"),
|
291 |
-
"reference": datasets.Value("string"),
|
292 |
-
}
|
293 |
-
)
|
294 |
-
|
295 |
-
return datasets.DatasetInfo(
|
296 |
-
# This is the description that will appear on the datasets page.
|
297 |
-
description=_DESCRIPTION,
|
298 |
-
# This defines the different columns of the dataset and their types
|
299 |
-
features=features, # Here we define them above because they are different between the two configurations
|
300 |
-
# If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
|
301 |
-
# specify them. They'll be used if as_supervised=True in builder.as_dataset.
|
302 |
-
# supervised_keys=("sentence", "label"),
|
303 |
-
# Homepage of the dataset for documentation
|
304 |
-
homepage=_HOMEPAGE,
|
305 |
-
# License for the dataset if available
|
306 |
-
license=_LICENSE,
|
307 |
-
# Citation for the dataset
|
308 |
-
citation=_CITATION,
|
309 |
-
)
|
310 |
-
|
311 |
-
def _post_process_dataframe(self, filepath):
|
312 |
-
def map_year(year):
|
313 |
-
if year <= 2017:
|
314 |
-
return "<=2017"
|
315 |
-
return str(year)
|
316 |
-
|
317 |
-
def normalize_grades(grades):
|
318 |
-
grades = grades.strip("[]").split(", ")
|
319 |
-
grade_mapping = {"0.0": 0, "20": 40, "2.0": 2}
|
320 |
-
|
321 |
-
# We will remove the rows that match the criteria below
|
322 |
-
if any(
|
323 |
-
single_grade
|
324 |
-
in grades[:-1] # we ignore the sum, and only check the concetps
|
325 |
-
for single_grade in ["50", "100", "150", "0.5", "1.0", "1.5"]
|
326 |
-
):
|
327 |
-
return None
|
328 |
-
# Use the mapping to transform grades, ignoring the last grade
|
329 |
-
mapped_grades = [
|
330 |
-
int(grade_mapping.get(grade_concept, grade_concept))
|
331 |
-
for grade_concept in grades[:-1]
|
332 |
-
]
|
333 |
-
# Calculate and append the sum of the mapped grades as the last element
|
334 |
-
mapped_grades.append(sum(mapped_grades))
|
335 |
-
return mapped_grades
|
336 |
-
|
337 |
-
df = pd.read_csv(filepath)
|
338 |
-
df["general"] = df["general"].fillna("")
|
339 |
-
df["essay_year"] = df["essay_year"].astype("int")
|
340 |
-
df["mapped_year"] = df["essay_year"].apply(map_year)
|
341 |
-
df["grades"] = df["grades"].apply(normalize_grades)
|
342 |
-
df = df.dropna(subset=["grades"])
|
343 |
-
df = df[
|
344 |
-
~(df["id_prompt"] + "/" + df["id"]).isin(ESSAY_TO_IGNORE)
|
345 |
-
] # arbitrary removal of zero graded essays
|
346 |
-
df.to_csv(filepath, index=False)
|
347 |
-
|
348 |
-
def _preprocess_propor2024(self, base_path: str):
|
349 |
-
for split_case in ["train.csv", "validation.csv", "test.csv"]:
|
350 |
-
filepath = f"{base_path}/propor2024/{split_case}"
|
351 |
-
df = pd.read_csv(filepath)
|
352 |
-
|
353 |
-
# Dictionary to track how many times we've seen each (id, id_prompt) pair
|
354 |
-
counts = {}
|
355 |
-
# List to store the reference for each row
|
356 |
-
references = []
|
357 |
-
|
358 |
-
# Define the mapping for each occurrence
|
359 |
-
occurrence_to_reference = {
|
360 |
-
0: "crawled_from_web",
|
361 |
-
1: "grader_a",
|
362 |
-
2: "grader_b",
|
363 |
-
}
|
364 |
-
|
365 |
-
# Iterate through rows in the original order
|
366 |
-
for _, row in df.iterrows():
|
367 |
-
key = (row["id"], row["id_prompt"])
|
368 |
-
count = counts.get(key, 0)
|
369 |
-
# Assign the reference based on the count
|
370 |
-
ref = occurrence_to_reference.get(count, "unknown")
|
371 |
-
references.append(ref)
|
372 |
-
counts[key] = count + 1
|
373 |
-
|
374 |
-
# Add the reference column without changing the order of rows
|
375 |
-
df["reference"] = references
|
376 |
-
df.to_csv(filepath, index=False)
|
377 |
-
|
378 |
-
def _split_generators(self, dl_manager):
|
379 |
-
if self.config.name != "JBCS2025":
|
380 |
-
urls = _URLS[self.config.name]
|
381 |
-
extracted_files = dl_manager.download_and_extract({self.config.name: urls})
|
382 |
-
if "PROPOR2024" == self.config.name:
|
383 |
-
base_path = extracted_files["PROPOR2024"]
|
384 |
-
self._preprocess_propor2024(base_path)
|
385 |
-
return [
|
386 |
-
datasets.SplitGenerator(
|
387 |
-
name=datasets.Split.TRAIN,
|
388 |
-
# These kwargs will be passed to _generate_examples
|
389 |
-
gen_kwargs={
|
390 |
-
"filepath": os.path.join(base_path, "propor2024/train.csv"),
|
391 |
-
"split": "train",
|
392 |
-
},
|
393 |
-
),
|
394 |
-
datasets.SplitGenerator(
|
395 |
-
name=datasets.Split.VALIDATION,
|
396 |
-
# These kwargs will be passed to _generate_examples
|
397 |
-
gen_kwargs={
|
398 |
-
"filepath": os.path.join(
|
399 |
-
base_path, "propor2024/validation.csv"
|
400 |
-
),
|
401 |
-
"split": "validation",
|
402 |
-
},
|
403 |
-
),
|
404 |
-
datasets.SplitGenerator(
|
405 |
-
name=datasets.Split.TEST,
|
406 |
-
gen_kwargs={
|
407 |
-
"filepath": os.path.join(base_path, "propor2024/test.csv"),
|
408 |
-
"split": "test",
|
409 |
-
},
|
410 |
-
),
|
411 |
-
]
|
412 |
-
if "gradesThousand" == self.config.name:
|
413 |
-
urls = _URLS[self.config.name]
|
414 |
-
extracted_files = dl_manager.download_and_extract({self.config.name: urls})
|
415 |
-
base_path = f"{extracted_files['gradesThousand']}/scrapedGradesThousand"
|
416 |
-
for split in ["train", "validation", "test"]:
|
417 |
-
split_filepath = os.path.join(base_path, f"{split}.csv")
|
418 |
-
grades_thousand = pd.read_csv(split_filepath)
|
419 |
-
grades_thousand[["supporting_text", "prompt"]] = grades_thousand[
|
420 |
-
"supporting_text"
|
421 |
-
].apply(
|
422 |
-
lambda original_text: pd.Series(
|
423 |
-
self._extract_prompt_and_clean(original_text)
|
424 |
-
)
|
425 |
-
)
|
426 |
-
grades_thousand.to_csv(split_filepath, index=False)
|
427 |
-
return [
|
428 |
-
datasets.SplitGenerator(
|
429 |
-
name=datasets.Split.TRAIN,
|
430 |
-
# These kwargs will be passed to _generate_examples
|
431 |
-
gen_kwargs={
|
432 |
-
"filepath": os.path.join(base_path, "train.csv"),
|
433 |
-
"split": "train",
|
434 |
-
},
|
435 |
-
),
|
436 |
-
datasets.SplitGenerator(
|
437 |
-
name=datasets.Split.VALIDATION,
|
438 |
-
# These kwargs will be passed to _generate_examples
|
439 |
-
gen_kwargs={
|
440 |
-
"filepath": os.path.join(base_path, "validation.csv"),
|
441 |
-
"split": "validation",
|
442 |
-
},
|
443 |
-
),
|
444 |
-
datasets.SplitGenerator(
|
445 |
-
name=datasets.Split.TEST,
|
446 |
-
gen_kwargs={
|
447 |
-
"filepath": os.path.join(base_path, "test.csv"),
|
448 |
-
"split": "test",
|
449 |
-
},
|
450 |
-
),
|
451 |
-
]
|
452 |
-
if "sourceA" in self.config.name:
|
453 |
-
html_parser = self._process_html_files(extracted_files)
|
454 |
-
self._post_process_dataframe(html_parser.sourceA)
|
455 |
-
self._generate_splits(html_parser.sourceA)
|
456 |
-
folder_sourceA = Path(html_parser.sourceA).parent
|
457 |
-
return [
|
458 |
-
datasets.SplitGenerator(
|
459 |
-
name=datasets.Split.TRAIN,
|
460 |
-
# These kwargs will be passed to _generate_examples
|
461 |
-
gen_kwargs={
|
462 |
-
"filepath": folder_sourceA / "train.csv",
|
463 |
-
"split": "train",
|
464 |
-
},
|
465 |
-
),
|
466 |
-
datasets.SplitGenerator(
|
467 |
-
name=datasets.Split.VALIDATION,
|
468 |
-
# These kwargs will be passed to _generate_examples
|
469 |
-
gen_kwargs={
|
470 |
-
"filepath": folder_sourceA / "validation.csv",
|
471 |
-
"split": "validation",
|
472 |
-
},
|
473 |
-
),
|
474 |
-
datasets.SplitGenerator(
|
475 |
-
name=datasets.Split.TEST,
|
476 |
-
gen_kwargs={
|
477 |
-
"filepath": folder_sourceA / "test.csv",
|
478 |
-
"split": "test",
|
479 |
-
},
|
480 |
-
),
|
481 |
-
]
|
482 |
-
elif self.config.name == "sourceB":
|
483 |
-
html_parser = self._process_html_files(extracted_files)
|
484 |
-
self._post_process_dataframe(html_parser.sourceB)
|
485 |
-
return [
|
486 |
-
datasets.SplitGenerator(
|
487 |
-
name="full",
|
488 |
-
gen_kwargs={
|
489 |
-
"filepath": html_parser.sourceB,
|
490 |
-
"split": "full",
|
491 |
-
},
|
492 |
-
),
|
493 |
-
]
|
494 |
-
elif "JBCS2025" == self.config.name:
|
495 |
-
extracted_files = dl_manager.download_and_extract(
|
496 |
-
{
|
497 |
-
"sourceA": _URLS["sourceAWithGraders"],
|
498 |
-
"grades_thousand": _URLS["gradesThousand"],
|
499 |
-
}
|
500 |
-
)
|
501 |
-
config_name_source_a = "sourceAWithGraders"
|
502 |
-
|
503 |
-
html_parser = self._process_html_files(
|
504 |
-
paths_dict={config_name_source_a: extracted_files["sourceA"]},
|
505 |
-
config_name=config_name_source_a,
|
506 |
-
)
|
507 |
-
self._post_process_dataframe(html_parser.sourceA)
|
508 |
-
self._generate_splits(html_parser.sourceA, config_name=config_name_source_a)
|
509 |
-
folder_sourceA = Path(html_parser.sourceA).parent
|
510 |
-
for split in ["train", "validation", "test"]:
|
511 |
-
sourceA = pd.read_csv(folder_sourceA / f"{split}.csv")
|
512 |
-
common_columns = [
|
513 |
-
"id",
|
514 |
-
"id_prompt",
|
515 |
-
"essay_text",
|
516 |
-
"grades",
|
517 |
-
"essay_year",
|
518 |
-
"supporting_text",
|
519 |
-
"prompt",
|
520 |
-
"reference",
|
521 |
-
]
|
522 |
-
combined_split = sourceA[
|
523 |
-
sourceA.reference.isin(["grader_a", "grader_b"])
|
524 |
-
]
|
525 |
-
combined_split = combined_split.rename(columns={"essay": "essay_text"})
|
526 |
-
combined_split["grades"] = combined_split["grades"].str.replace(",", "")
|
527 |
-
final_split = combined_split[common_columns].sample(
|
528 |
-
frac=1, random_state=RANDOM_STATE
|
529 |
-
).reset_index(drop=True)
|
530 |
-
# overwrites the sourceA data
|
531 |
-
final_split.to_csv(folder_sourceA / f"{split}.csv", index=False)
|
532 |
-
return [
|
533 |
-
datasets.SplitGenerator(
|
534 |
-
name=datasets.Split.TRAIN,
|
535 |
-
# These kwargs will be passed to _generate_examples
|
536 |
-
gen_kwargs={
|
537 |
-
"filepath": folder_sourceA / "train.csv",
|
538 |
-
"split": "train",
|
539 |
-
},
|
540 |
-
),
|
541 |
-
datasets.SplitGenerator(
|
542 |
-
name=datasets.Split.VALIDATION,
|
543 |
-
# These kwargs will be passed to _generate_examples
|
544 |
-
gen_kwargs={
|
545 |
-
"filepath": folder_sourceA / "validation.csv",
|
546 |
-
"split": "validation",
|
547 |
-
},
|
548 |
-
),
|
549 |
-
datasets.SplitGenerator(
|
550 |
-
name=datasets.Split.TEST,
|
551 |
-
gen_kwargs={
|
552 |
-
"filepath": folder_sourceA / "test.csv",
|
553 |
-
"split": "test",
|
554 |
-
},
|
555 |
-
),
|
556 |
-
]
|
557 |
-
|
558 |
-
def _extract_prompt_and_clean(self, text: str):
|
559 |
-
"""
|
560 |
-
1) Find an uppercase block matching "PROPOSTA DE REDACAO/REDAÇÃO"
|
561 |
-
(with flexible spacing and accents) anywhere in 'text'.
|
562 |
-
2) Capture everything from there until the next heading that
|
563 |
-
starts a line (TEXTO..., TEXTOS..., INSTRUÇÕES...) or end-of-text.
|
564 |
-
3) Remove that captured block from the original, returning:
|
565 |
-
(supporting_text, prompt)
|
566 |
-
"""
|
567 |
-
|
568 |
-
# Regex explanation:
|
569 |
-
# (?m) => MULTILINE, so ^ can match start of lines
|
570 |
-
# 1) PROPOSTA\s+DE\s+REDA(?:C|Ç)(?:AO|ÃO)
|
571 |
-
# - "PROPOSTA", then one-or-more spaces/newlines,
|
572 |
-
# then "DE", then spaces, then "REDA(C|Ç)",
|
573 |
-
# and either "AO" or "ÃO" (uppercase).
|
574 |
-
# - This part may skip diacritic or accent variations in "REDAÇÃO" vs. "REDACAO".
|
575 |
-
#
|
576 |
-
# 2) (?:.*?\n?)*? => a non-greedy capture of subsequent lines
|
577 |
-
# (including possible newlines). We use [\s\S]*? as an alternative.
|
578 |
-
#
|
579 |
-
# 3) Lookahead (?=^(?:TEXTO|TEXTOS|INSTRUÇÕES|\Z))
|
580 |
-
# means: stop right before a line that starts with "TEXTO", "TEXTOS",
|
581 |
-
# or "INSTRUÇÕES", OR the very end of the text (\Z).
|
582 |
-
#
|
583 |
-
# If found, that entire portion is group(1).
|
584 |
-
def force_newline_after_proposta(text: str) -> str:
|
585 |
-
"""
|
586 |
-
If we see "PROPOSTA DE REDAÇÃO" immediately followed by some
|
587 |
-
non-whitespace character (like "A"), insert two newlines.
|
588 |
-
E.g., "PROPOSTA DE REDAÇÃOA partir..." becomes
|
589 |
-
"PROPOSTA DE REDAÇÃO\n\nA partir..."
|
590 |
-
"""
|
591 |
-
# This pattern looks for:
|
592 |
-
# (PROPOSTA DE REDAÇÃO)
|
593 |
-
# (?=\S) meaning "immediately followed by a NON-whitespace character"
|
594 |
-
# then we replace that with "PROPOSTA DE REDAÇÃO\n\n"
|
595 |
-
pattern = re.compile(r"(?=\S)(PROPOSTA DE REDAÇÃO)(?=\S)")
|
596 |
-
return pattern.sub(r"\n\1\n\n", text)
|
597 |
-
|
598 |
-
text = force_newline_after_proposta(text)
|
599 |
-
pattern = re.compile(
|
600 |
-
r"(?m)" # MULTILINE
|
601 |
-
r"("
|
602 |
-
r"PROPOSTA\s+DE\s+REDA(?:C|Ç)(?:AO|ÃO)" # e.g. PROPOSTA DE REDACAO / REDAÇÃO
|
603 |
-
r"(?:[\s\S]*?)" # lazily grab the subsequent text
|
604 |
-
r")"
|
605 |
-
r"(?=(?:TEXTO|TEXTOS|INSTRUÇÕES|TExTO|\Z))"
|
606 |
-
)
|
607 |
-
|
608 |
-
match = pattern.search(text)
|
609 |
-
if match:
|
610 |
-
prompt = match.group(1).strip()
|
611 |
-
# Remove that block from the original:
|
612 |
-
start, end = match.span(1)
|
613 |
-
main_text = text[:start] + text[end:]
|
614 |
-
else:
|
615 |
-
# No match => keep entire text in supporting_text, prompt empty
|
616 |
-
prompt = ""
|
617 |
-
main_text = text
|
618 |
-
|
619 |
-
return main_text.strip(), prompt.strip()
|
620 |
-
|
621 |
-
def _process_html_files(self, paths_dict, config_name=None):
|
622 |
-
html_parser = HTMLParser(paths_dict)
|
623 |
-
if config_name is None:
|
624 |
-
config_name = self.config.name
|
625 |
-
html_parser.parse(config_name)
|
626 |
-
return html_parser
|
627 |
-
|
628 |
-
def _parse_graders_data(self, dirname):
|
629 |
-
map_grades = {"0": 0, "1": 40, "2": 80, "3": 120, "4": 160, "5": 200}
|
630 |
-
|
631 |
-
def map_list(grades_list):
|
632 |
-
result = [map_grades.get(item, None) for item in grades_list]
|
633 |
-
sum_grades = sum(result)
|
634 |
-
result.append(sum_grades)
|
635 |
-
return result
|
636 |
-
|
637 |
-
grader_a = pd.read_csv(f"{dirname}/GraderA.csv")
|
638 |
-
grader_b = pd.read_csv(f"{dirname}/GraderB.csv")
|
639 |
-
for grader in [grader_a, grader_b]:
|
640 |
-
grader.grades = grader.grades.apply(lambda x: x.strip("[]").split(", "))
|
641 |
-
grader.grades = grader.grades.apply(map_list)
|
642 |
-
grader_a["reference"] = "grader_a"
|
643 |
-
grader_b["reference"] = "grader_b"
|
644 |
-
return grader_a, grader_b
|
645 |
-
|
646 |
-
def _generate_splits(self, filepath: str, train_size=0.7, config_name=None):
|
647 |
-
np.random.seed(RANDOM_STATE)
|
648 |
-
df = pd.read_csv(filepath)
|
649 |
-
train_set = []
|
650 |
-
val_set = []
|
651 |
-
test_set = []
|
652 |
-
df = df.sort_values(by=["essay_year", "id_prompt"]).reset_index(drop=True)
|
653 |
-
buckets = {}
|
654 |
-
for key, group in df.groupby("mapped_year"):
|
655 |
-
buckets[key] = sorted(group["id_prompt"].unique())
|
656 |
-
df.drop("mapped_year", axis=1, inplace=True)
|
657 |
-
for year in sorted(buckets.keys()):
|
658 |
-
prompts = buckets[year]
|
659 |
-
np.random.shuffle(prompts)
|
660 |
-
num_prompts = len(prompts)
|
661 |
-
|
662 |
-
# All prompts go to the test if less than 3
|
663 |
-
if num_prompts <= 3:
|
664 |
-
train_set.append(df[df["id_prompt"].isin([prompts[0]])])
|
665 |
-
val_set.append(df[df["id_prompt"].isin([prompts[1]])])
|
666 |
-
test_set.append(df[df["id_prompt"].isin([prompts[2]])])
|
667 |
-
continue
|
668 |
-
|
669 |
-
# Determine the number of prompts for each set based on train_size and remaining prompts
|
670 |
-
num_train = math.floor(num_prompts * train_size)
|
671 |
-
num_val_test = num_prompts - num_train
|
672 |
-
num_val = num_val_test // 2
|
673 |
-
num_test = num_val_test - num_val
|
674 |
-
|
675 |
-
# Assign prompts to each set
|
676 |
-
train_set.append(df[df["id_prompt"].isin(prompts[:num_train])])
|
677 |
-
val_set.append(
|
678 |
-
df[df["id_prompt"].isin(prompts[num_train : (num_train + num_val)])]
|
679 |
-
)
|
680 |
-
test_set.append(
|
681 |
-
df[
|
682 |
-
df["id_prompt"].isin(
|
683 |
-
prompts[
|
684 |
-
(num_train + num_val) : (num_train + num_val + num_test)
|
685 |
-
]
|
686 |
-
)
|
687 |
-
]
|
688 |
-
)
|
689 |
-
|
690 |
-
# Convert lists of groups to DataFrames
|
691 |
-
train_df = pd.concat(train_set)
|
692 |
-
val_df = pd.concat(val_set)
|
693 |
-
test_df = pd.concat(test_set)
|
694 |
-
dirname = os.path.dirname(filepath)
|
695 |
-
if config_name is None:
|
696 |
-
config_name = self.config.name
|
697 |
-
if config_name == "sourceAWithGraders":
|
698 |
-
grader_a, grader_b = self._parse_graders_data(dirname)
|
699 |
-
grader_a_data = pd.merge(
|
700 |
-
train_df[["id", "id_prompt", "essay", "prompt", "supporting_text"]],
|
701 |
-
grader_a.drop(columns=["essay"]),
|
702 |
-
on=["id", "id_prompt"],
|
703 |
-
how="inner",
|
704 |
-
)
|
705 |
-
grader_b_data = pd.merge(
|
706 |
-
train_df[["id", "id_prompt", "essay", "prompt", "supporting_text"]],
|
707 |
-
grader_b.drop(columns=["essay"]),
|
708 |
-
on=["id", "id_prompt"],
|
709 |
-
how="inner",
|
710 |
-
)
|
711 |
-
train_df = pd.concat([train_df, grader_a_data, grader_b_data])
|
712 |
-
train_df = train_df.sort_values(by=["id", "id_prompt"]).reset_index(
|
713 |
-
drop=True
|
714 |
-
)
|
715 |
-
|
716 |
-
grader_a_data = pd.merge(
|
717 |
-
val_df[["id", "id_prompt", "essay", "prompt", "supporting_text"]],
|
718 |
-
grader_a.drop(columns=["essay"]),
|
719 |
-
on=["id", "id_prompt"],
|
720 |
-
how="inner",
|
721 |
-
)
|
722 |
-
grader_b_data = pd.merge(
|
723 |
-
val_df[["id", "id_prompt", "essay", "prompt", "supporting_text"]],
|
724 |
-
grader_b.drop(columns=["essay"]),
|
725 |
-
on=["id", "id_prompt"],
|
726 |
-
how="inner",
|
727 |
-
)
|
728 |
-
val_df = pd.concat([val_df, grader_a_data, grader_b_data])
|
729 |
-
val_df = val_df.sort_values(by=["id", "id_prompt"]).reset_index(drop=True)
|
730 |
-
|
731 |
-
grader_a_data = pd.merge(
|
732 |
-
test_df[["id", "id_prompt", "essay", "prompt", "supporting_text"]],
|
733 |
-
grader_a.drop(columns=["essay"]),
|
734 |
-
on=["id", "id_prompt"],
|
735 |
-
how="inner",
|
736 |
-
)
|
737 |
-
grader_b_data = pd.merge(
|
738 |
-
test_df[["id", "id_prompt", "essay", "prompt", "supporting_text"]],
|
739 |
-
grader_b.drop(columns=["essay"]),
|
740 |
-
on=["id", "id_prompt"],
|
741 |
-
how="inner",
|
742 |
-
)
|
743 |
-
test_df = pd.concat([test_df, grader_a_data, grader_b_data])
|
744 |
-
test_df = test_df.sort_values(by=["id", "id_prompt"]).reset_index(drop=True)
|
745 |
-
|
746 |
-
train_df = train_df.sample(frac=1, random_state=RANDOM_STATE).reset_index(
|
747 |
-
drop=True
|
748 |
-
)
|
749 |
-
val_df = val_df.sample(frac=1, random_state=RANDOM_STATE).reset_index(
|
750 |
-
drop=True
|
751 |
-
)
|
752 |
-
test_df = test_df.sample(frac=1, random_state=RANDOM_STATE).reset_index(
|
753 |
-
drop=True
|
754 |
-
)
|
755 |
-
|
756 |
-
# Data Validation Assertions
|
757 |
-
assert (
|
758 |
-
len(set(train_df["id_prompt"]).intersection(set(val_df["id_prompt"]))) == 0
|
759 |
-
), "Overlap between train and val id_prompt"
|
760 |
-
assert (
|
761 |
-
len(set(train_df["id_prompt"]).intersection(set(test_df["id_prompt"]))) == 0
|
762 |
-
), "Overlap between train and test id_prompt"
|
763 |
-
assert (
|
764 |
-
len(set(val_df["id_prompt"]).intersection(set(test_df["id_prompt"]))) == 0
|
765 |
-
), "Overlap between val and test id_prompt"
|
766 |
-
train_df.to_csv(f"{dirname}/train.csv", index=False)
|
767 |
-
val_df.to_csv(f"{dirname}/validation.csv", index=False)
|
768 |
-
test_df.to_csv(f"{dirname}/test.csv", index=False)
|
769 |
-
|
770 |
-
# method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
|
771 |
-
def _generate_examples(self, filepath, split):
|
772 |
-
if self.config.name == "PROPOR2024":
|
773 |
-
with open(filepath, encoding="utf-8") as csvfile:
|
774 |
-
next(csvfile)
|
775 |
-
csv_reader = csv.DictReader(csvfile, fieldnames=CSV_HEADERPROPOR)
|
776 |
-
for i, row in enumerate(csv_reader):
|
777 |
-
grades = row["grades"].strip("[]")
|
778 |
-
grades = grades.split()
|
779 |
-
yield (
|
780 |
-
i,
|
781 |
-
{
|
782 |
-
"id": row["id"],
|
783 |
-
"id_prompt": row["id_prompt"],
|
784 |
-
"essay_title": row["title"],
|
785 |
-
"essay_text": row["essay"],
|
786 |
-
"grades": grades,
|
787 |
-
"essay_year": row["essay_year"],
|
788 |
-
"reference": row["reference"],
|
789 |
-
},
|
790 |
-
)
|
791 |
-
elif self.config.name == "gradesThousand":
|
792 |
-
with open(filepath, encoding="utf-8") as csvfile:
|
793 |
-
next(csvfile)
|
794 |
-
csv_reader = csv.DictReader(csvfile, fieldnames=CSV_HEADERTHOUSAND)
|
795 |
-
for i, row in enumerate(csv_reader):
|
796 |
-
grades = row["grades"].strip("[]")
|
797 |
-
grades = grades.split(", ")
|
798 |
-
yield (
|
799 |
-
i,
|
800 |
-
{
|
801 |
-
"id": row["id"],
|
802 |
-
"id_prompt": row["id_prompt"],
|
803 |
-
"supporting_text": row["supporting_text"],
|
804 |
-
"prompt": row["prompt"],
|
805 |
-
"essay_text": row["essay"],
|
806 |
-
"grades": grades,
|
807 |
-
"essay_year": row["essay_year"],
|
808 |
-
"author": row["author"],
|
809 |
-
"source": row["source"],
|
810 |
-
},
|
811 |
-
)
|
812 |
-
elif self.config.name == "JBCS2025":
|
813 |
-
with open(filepath, encoding="utf-8") as csvfile:
|
814 |
-
next(csvfile)
|
815 |
-
csv_reader = csv.DictReader(csvfile, fieldnames=CSV_HEADER_JBCS25)
|
816 |
-
for i, row in enumerate(csv_reader):
|
817 |
-
grades = row["grades"].strip("[]")
|
818 |
-
grades = grades.split()
|
819 |
-
yield (
|
820 |
-
i,
|
821 |
-
{
|
822 |
-
"id": row["id"],
|
823 |
-
"id_prompt": row["id_prompt"],
|
824 |
-
"essay_text": row["essay_text"],
|
825 |
-
"grades": grades,
|
826 |
-
"essay_year": row["essay_year"],
|
827 |
-
"supporting_text": row["supporting_text"],
|
828 |
-
"prompt": row["prompt"],
|
829 |
-
"reference": row["reference"],
|
830 |
-
},
|
831 |
-
)
|
832 |
-
else:
|
833 |
-
with open(filepath, encoding="utf-8") as csvfile:
|
834 |
-
next(csvfile)
|
835 |
-
csv_reader = csv.DictReader(csvfile, fieldnames=CSV_HEADER)
|
836 |
-
for i, row in enumerate(csv_reader):
|
837 |
-
grades = row["grades"].strip("[]")
|
838 |
-
grades = grades.split(", ")
|
839 |
-
yield (
|
840 |
-
i,
|
841 |
-
{
|
842 |
-
"id": row["id"],
|
843 |
-
"id_prompt": row["id_prompt"],
|
844 |
-
"prompt": row["prompt"],
|
845 |
-
"supporting_text": row["supporting_text"],
|
846 |
-
"essay_title": row["title"],
|
847 |
-
"essay_text": row["essay"],
|
848 |
-
"grades": grades,
|
849 |
-
"essay_year": row["essay_year"],
|
850 |
-
"general_comment": row["general"],
|
851 |
-
"specific_comment": row["specific"],
|
852 |
-
"reference": row["reference"],
|
853 |
-
},
|
854 |
-
)
|
855 |
-
|
856 |
-
|
857 |
-
class HTMLParser:
|
858 |
-
def __init__(self, paths_dict):
|
859 |
-
self.paths_dict = paths_dict
|
860 |
-
self.sourceA = None
|
861 |
-
self.sourceB = None
|
862 |
-
|
863 |
-
def apply_soup(self, filepath, num):
|
864 |
-
# recebe uma URL, salva o HTML dessa página e retorna o soup dela
|
865 |
-
file = open(os.path.join(filepath, num), "r", encoding="utf8")
|
866 |
-
conteudo = file.read()
|
867 |
-
soup = BeautifulSoup(conteudo, "html.parser")
|
868 |
-
return soup
|
869 |
-
|
870 |
-
def _get_title(self, soup):
|
871 |
-
if self.sourceA:
|
872 |
-
title = soup.find("div", class_="container-composition")
|
873 |
-
if title is None:
|
874 |
-
title = soup.find("h1", class_="pg-color10").get_text()
|
875 |
-
else:
|
876 |
-
title = title.h2.get_text()
|
877 |
-
title = title.replace("\xa0", "")
|
878 |
-
return title.replace(";", ",")
|
879 |
-
elif self.sourceB:
|
880 |
-
title = soup.find("h1", class_="titulo-conteudo").get_text()
|
881 |
-
return title.strip("- Banco de redações").strip()
|
882 |
-
|
883 |
-
def _get_grades(self, soup):
|
884 |
-
if self.sourceA:
|
885 |
-
grades = soup.find("section", class_="results-table")
|
886 |
-
final_grades = []
|
887 |
-
if grades is not None:
|
888 |
-
grades = grades.find_all("span", class_="points")
|
889 |
-
assert len(grades) == 6, f"Missing grades: {len(grades)}"
|
890 |
-
for single_grade in grades:
|
891 |
-
grade = int(single_grade.get_text())
|
892 |
-
final_grades.append(grade)
|
893 |
-
assert final_grades[-1] == sum(final_grades[:-1]), (
|
894 |
-
"Grading sum is not making sense"
|
895 |
-
)
|
896 |
-
else:
|
897 |
-
grades = soup.find("div", class_="redacoes-corrigidas pg-bordercolor7")
|
898 |
-
grades_sum = float(
|
899 |
-
soup.find("th", class_="noBorder-left").get_text().replace(",", ".")
|
900 |
-
)
|
901 |
-
grades = grades.find_all("td")[:10]
|
902 |
-
for idx in range(1, 10, 2):
|
903 |
-
grade = float(grades[idx].get_text().replace(",", "."))
|
904 |
-
final_grades.append(grade)
|
905 |
-
assert grades_sum == sum(final_grades), (
|
906 |
-
"Grading sum is not making sense"
|
907 |
-
)
|
908 |
-
final_grades.append(grades_sum)
|
909 |
-
return final_grades
|
910 |
-
elif self.sourceB:
|
911 |
-
table = soup.find("table", {"id": "redacoes_corrigidas"})
|
912 |
-
grades = table.find_all("td", class_="simple-td")
|
913 |
-
grades = grades[3:]
|
914 |
-
result = []
|
915 |
-
for single_grade in grades:
|
916 |
-
result.append(int(single_grade.get_text()))
|
917 |
-
assert len(result) == 5, "We should have 5 Grades (one per concept) only"
|
918 |
-
result.append(
|
919 |
-
sum(result)
|
920 |
-
) # Add sum as a sixt element to keep the same pattern
|
921 |
-
return result
|
922 |
-
|
923 |
-
def _get_general_comment(self, soup):
|
924 |
-
if self.sourceA:
|
925 |
-
|
926 |
-
def get_general_comment_aux(soup):
|
927 |
-
result = soup.find("article", class_="list-item c")
|
928 |
-
if result is not None:
|
929 |
-
result = result.find("div", class_="description")
|
930 |
-
return result.get_text()
|
931 |
-
else:
|
932 |
-
result = soup.find("p", style="margin: 0px 0px 11px;")
|
933 |
-
if result is not None:
|
934 |
-
return result.get_text()
|
935 |
-
else:
|
936 |
-
result = soup.find("p", style="margin: 0px;")
|
937 |
-
if result is not None:
|
938 |
-
return result.get_text()
|
939 |
-
else:
|
940 |
-
result = soup.find(
|
941 |
-
"p", style="margin: 0px; text-align: justify;"
|
942 |
-
)
|
943 |
-
if result is not None:
|
944 |
-
return result.get_text()
|
945 |
-
else:
|
946 |
-
return ""
|
947 |
-
|
948 |
-
text = soup.find("div", class_="text")
|
949 |
-
if text is not None:
|
950 |
-
text = text.find("p")
|
951 |
-
if (text is None) or (len(text.get_text()) < 2):
|
952 |
-
return get_general_comment_aux(soup)
|
953 |
-
return text.get_text()
|
954 |
-
else:
|
955 |
-
return get_general_comment_aux(soup)
|
956 |
-
elif self.sourceB:
|
957 |
-
return ""
|
958 |
-
|
959 |
-
def _get_specific_comment(self, soup, general_comment):
|
960 |
-
if self.sourceA:
|
961 |
-
result = soup.find("div", class_="text")
|
962 |
-
cms = []
|
963 |
-
if result is not None:
|
964 |
-
result = result.find_all("li")
|
965 |
-
if result != []:
|
966 |
-
for item in result:
|
967 |
-
text = item.get_text()
|
968 |
-
if text != "\xa0":
|
969 |
-
cms.append(text)
|
970 |
-
else:
|
971 |
-
result = soup.find("div", class_="text").find_all("p")
|
972 |
-
for item in result:
|
973 |
-
text = item.get_text()
|
974 |
-
if text != "\xa0":
|
975 |
-
cms.append(text)
|
976 |
-
else:
|
977 |
-
result = soup.find_all("article", class_="list-item c")
|
978 |
-
if len(result) < 2:
|
979 |
-
return ["First if"]
|
980 |
-
result = result[1].find_all("p")
|
981 |
-
for item in result:
|
982 |
-
text = item.get_text()
|
983 |
-
if text != "\xa0":
|
984 |
-
cms.append(text)
|
985 |
-
specific_comment = cms.copy()
|
986 |
-
if general_comment in specific_comment:
|
987 |
-
specific_comment.remove(general_comment)
|
988 |
-
if (len(specific_comment) > 1) and (len(specific_comment[0]) < 2):
|
989 |
-
specific_comment = specific_comment[1:]
|
990 |
-
return self._clean_list(specific_comment)
|
991 |
-
elif self.sourceB:
|
992 |
-
return ""
|
993 |
-
|
994 |
-
def _get_essay(self, soup):
|
995 |
-
if self.sourceA:
|
996 |
-
essay = soup.find("div", class_="text-composition")
|
997 |
-
result = []
|
998 |
-
if essay is not None:
|
999 |
-
essay = essay.find_all("p")
|
1000 |
-
for f in essay:
|
1001 |
-
while f.find("span", style="color:#00b050") is not None:
|
1002 |
-
f.find("span", style="color:#00b050").decompose()
|
1003 |
-
while f.find("span", class_="certo") is not None:
|
1004 |
-
f.find("span", class_="certo").decompose()
|
1005 |
-
for paragraph in essay:
|
1006 |
-
result.append(paragraph.get_text())
|
1007 |
-
else:
|
1008 |
-
essay = soup.find("div", {"id": "texto"})
|
1009 |
-
essay.find("section", class_="list-items").decompose()
|
1010 |
-
essay = essay.find_all("p")
|
1011 |
-
for f in essay:
|
1012 |
-
while f.find("span", class_="certo") is not None:
|
1013 |
-
f.find("span", class_="certo").decompose()
|
1014 |
-
for paragraph in essay:
|
1015 |
-
result.append(paragraph.get_text())
|
1016 |
-
return "\n".join(self._clean_list(result))
|
1017 |
-
elif self.sourceB:
|
1018 |
-
table = soup.find("article", class_="texto-conteudo entire")
|
1019 |
-
table = soup.find("div", class_="area-redacao-corrigida")
|
1020 |
-
if table is None:
|
1021 |
-
result = None
|
1022 |
-
else:
|
1023 |
-
for span in soup.find_all("span"):
|
1024 |
-
span.decompose()
|
1025 |
-
result = table.find_all("p")
|
1026 |
-
result = " ".join(
|
1027 |
-
[
|
1028 |
-
paragraph.get_text().replace("\xa0", "").strip()
|
1029 |
-
for paragraph in result
|
1030 |
-
]
|
1031 |
-
)
|
1032 |
-
return result
|
1033 |
-
|
1034 |
-
def _get_essay_year(self, soup):
|
1035 |
-
if self.sourceA:
|
1036 |
-
pattern = r"redações corrigidas - \w+/\d+"
|
1037 |
-
first_occurrence = re.search(pattern, soup.get_text().lower())
|
1038 |
-
matched_url = first_occurrence.group(0) if first_occurrence else None
|
1039 |
-
year_pattern = r"\d{4}"
|
1040 |
-
return re.search(year_pattern, matched_url).group(0)
|
1041 |
-
elif self.sourceB:
|
1042 |
-
pattern = r"Enviou seu texto em.*?(\d{4})"
|
1043 |
-
match = re.search(pattern, soup.get_text())
|
1044 |
-
return match.group(1) if match else -1
|
1045 |
-
|
1046 |
-
def _clean_title(self, title):
|
1047 |
-
if self.sourceA:
|
1048 |
-
smaller_index = title.find("[")
|
1049 |
-
if smaller_index == -1:
|
1050 |
-
return title
|
1051 |
-
else:
|
1052 |
-
bigger_index = title.find("]")
|
1053 |
-
new_title = title[:smaller_index] + title[bigger_index + 1 :]
|
1054 |
-
return self._clean_title(new_title.replace(" ", " "))
|
1055 |
-
elif self.sourceB:
|
1056 |
-
return title
|
1057 |
-
|
1058 |
-
def _clean_list(self, list):
|
1059 |
-
if list == []:
|
1060 |
-
return []
|
1061 |
-
else:
|
1062 |
-
new_list = []
|
1063 |
-
for phrase in list:
|
1064 |
-
phrase = (
|
1065 |
-
phrase.replace("\xa0", "").replace(" ,", ",").replace(" .", ".")
|
1066 |
-
)
|
1067 |
-
while phrase.find(" ") != -1:
|
1068 |
-
phrase = phrase.replace(" ", " ")
|
1069 |
-
if len(phrase) > 1:
|
1070 |
-
new_list.append(phrase)
|
1071 |
-
return new_list
|
1072 |
-
|
1073 |
-
def _clean_string(self, sentence):
|
1074 |
-
sentence = sentence.replace("\xa0", "").replace("\u200b", "")
|
1075 |
-
sentence = (
|
1076 |
-
sentence.replace(".", ". ")
|
1077 |
-
.replace("?", "? ")
|
1078 |
-
.replace("!", "! ")
|
1079 |
-
.replace(")", ") ")
|
1080 |
-
.replace(":", ": ")
|
1081 |
-
.replace("”", "” ")
|
1082 |
-
)
|
1083 |
-
sentence = sentence.replace(" ", " ").replace(". . . ", "...")
|
1084 |
-
sentence = sentence.replace("(editado)", "").replace("(Editado)", "")
|
1085 |
-
sentence = sentence.replace("(editado e adaptado)", "").replace(
|
1086 |
-
"(Editado e adaptado)", ""
|
1087 |
-
)
|
1088 |
-
sentence = sentence.replace(". com. br", ".com.br")
|
1089 |
-
sentence = sentence.replace("[Veja o texto completo aqui]", "")
|
1090 |
-
return sentence
|
1091 |
-
|
1092 |
-
def _get_supporting_text(self, soup):
|
1093 |
-
if self.sourceA:
|
1094 |
-
textos = soup.find_all("ul", class_="article-wording-item")
|
1095 |
-
resposta = []
|
1096 |
-
for t in textos[:-1]:
|
1097 |
-
resposta.append(
|
1098 |
-
t.find("h3", class_="item-titulo").get_text().replace("\xa0", "")
|
1099 |
-
)
|
1100 |
-
resposta.append(
|
1101 |
-
self._clean_string(
|
1102 |
-
t.find("div", class_="item-descricao").get_text()
|
1103 |
-
)
|
1104 |
-
)
|
1105 |
-
return resposta
|
1106 |
-
else:
|
1107 |
-
return ""
|
1108 |
-
|
1109 |
-
def _get_prompt(self, soup):
|
1110 |
-
if self.sourceA:
|
1111 |
-
prompt = soup.find("div", class_="text").find_all("p")
|
1112 |
-
if len(prompt[0].get_text()) < 2:
|
1113 |
-
return [prompt[1].get_text().replace("\xa0", "")]
|
1114 |
-
else:
|
1115 |
-
return [prompt[0].get_text().replace("\xa0", "")]
|
1116 |
-
else:
|
1117 |
-
return ""
|
1118 |
-
|
1119 |
-
def _process_all_prompts(self, sub_folders, file_dir, reference, prompts_to_ignore):
|
1120 |
-
"""
|
1121 |
-
Process all prompt folders in parallel and return all rows to write.
|
1122 |
-
|
1123 |
-
Args:
|
1124 |
-
sub_folders (list): List of prompt folder names (or Paths).
|
1125 |
-
file_dir (str): Base directory where prompts are located.
|
1126 |
-
reference: Reference info to include in each row.
|
1127 |
-
prompts_to_ignore (collection): Prompts to be ignored.
|
1128 |
-
|
1129 |
-
Returns:
|
1130 |
-
list: A list of all rows to write to the CSV.
|
1131 |
-
"""
|
1132 |
-
|
1133 |
-
args_list = [
|
1134 |
-
(prompt_folder, file_dir, reference, prompts_to_ignore, self)
|
1135 |
-
for prompt_folder in sub_folders
|
1136 |
-
]
|
1137 |
-
|
1138 |
-
all_rows = []
|
1139 |
-
# Use a Pool to parallelize processing.
|
1140 |
-
with Pool(processes=cpu_count()) as pool:
|
1141 |
-
# Using imap allows us to update the progress bar.
|
1142 |
-
for rows in tqdm(
|
1143 |
-
pool.imap(HTMLParser._process_prompt_folder, args_list),
|
1144 |
-
total=len(args_list),
|
1145 |
-
desc="Processing prompts",
|
1146 |
-
):
|
1147 |
-
all_rows.extend(rows)
|
1148 |
-
return all_rows
|
1149 |
-
|
1150 |
-
def parse(self, config_name: str):
|
1151 |
-
for key, filepath in self.paths_dict.items():
|
1152 |
-
if key != config_name:
|
1153 |
-
continue # TODO improve later, we will only support a single config at a time
|
1154 |
-
if "sourceA" in config_name:
|
1155 |
-
self.sourceA = f"{filepath}/sourceA/sourceA.csv"
|
1156 |
-
elif config_name == "sourceB":
|
1157 |
-
self.sourceB = f"{filepath}/sourceB/sourceB.csv"
|
1158 |
-
file = self.sourceA if self.sourceA else self.sourceB
|
1159 |
-
file_path = Path(file)
|
1160 |
-
file_dir = file_path.parent
|
1161 |
-
sorted_files = sorted(file_dir.iterdir(), key=lambda p: p.name)
|
1162 |
-
sub_folders = [name for name in sorted_files if name.suffix != ".csv"]
|
1163 |
-
reference = "crawled_from_web"
|
1164 |
-
all_rows = self._process_all_prompts(
|
1165 |
-
sub_folders, file_dir, reference, PROMPTS_TO_IGNORE
|
1166 |
-
)
|
1167 |
-
with open(file_path, "w", newline="", encoding="utf8") as final_file:
|
1168 |
-
writer = csv.writer(final_file)
|
1169 |
-
writer.writerow(CSV_HEADER)
|
1170 |
-
for row in all_rows:
|
1171 |
-
writer.writerow(row)
|
1172 |
-
|
1173 |
-
@staticmethod
|
1174 |
-
def _process_prompt_folder(args):
|
1175 |
-
"""
|
1176 |
-
Process one prompt folder and return a list of rows to write to CSV.
|
1177 |
-
Args:
|
1178 |
-
args (tuple): Contains:
|
1179 |
-
- prompt_folder: The folder name (or Path object) for the prompt.
|
1180 |
-
- file_dir: The base directory.
|
1181 |
-
- reference: Reference info to include in each row.
|
1182 |
-
- prompts_to_ignore: A collection of prompts to skip.
|
1183 |
-
- instance: An instance of the class that contains the parsing methods.
|
1184 |
-
Returns:
|
1185 |
-
list: A list of rows (each row is a list) to write to CSV.
|
1186 |
-
"""
|
1187 |
-
prompt_folder, file_dir, reference, prompts_to_ignore, instance = args
|
1188 |
-
rows = []
|
1189 |
-
# Skip folders that should be ignored.
|
1190 |
-
if prompt_folder in prompts_to_ignore:
|
1191 |
-
return rows
|
1192 |
-
# Build the full path for the prompt folder.
|
1193 |
-
prompt = os.path.join(file_dir, prompt_folder)
|
1194 |
-
# List and sort the HTML files.
|
1195 |
-
try:
|
1196 |
-
sorted_prompts = sorted(os.listdir(prompt))
|
1197 |
-
except Exception as e:
|
1198 |
-
print(f"Error listing directory {prompt}: {e}")
|
1199 |
-
return rows
|
1200 |
-
# Process the common "Prompt.html" once.
|
1201 |
-
soup_prompt = instance.apply_soup(prompt, "Prompt.html")
|
1202 |
-
essay_year = instance._get_essay_year(soup_prompt)
|
1203 |
-
essay_supporting_text = "\n".join(instance._get_supporting_text(soup_prompt))
|
1204 |
-
essay_prompt = "\n".join(instance._get_prompt(soup_prompt))
|
1205 |
-
# Process each essay file except the prompt itself.
|
1206 |
-
for essay_filename in sorted_prompts:
|
1207 |
-
if essay_filename == "Prompt.html":
|
1208 |
-
continue
|
1209 |
-
soup_text = instance.apply_soup(prompt, essay_filename)
|
1210 |
-
essay_title = instance._clean_title(instance._get_title(soup_text))
|
1211 |
-
essay_grades = instance._get_grades(soup_text)
|
1212 |
-
essay_text = instance._get_essay(soup_text)
|
1213 |
-
general_comment = instance._get_general_comment(soup_text).strip()
|
1214 |
-
specific_comment = instance._get_specific_comment(
|
1215 |
-
soup_text, general_comment
|
1216 |
-
)
|
1217 |
-
# Create a row with all the information.
|
1218 |
-
row = [
|
1219 |
-
essay_filename,
|
1220 |
-
prompt_folder
|
1221 |
-
if not hasattr(prompt_folder, "name")
|
1222 |
-
else prompt_folder.name,
|
1223 |
-
essay_prompt,
|
1224 |
-
essay_supporting_text,
|
1225 |
-
essay_title,
|
1226 |
-
essay_text,
|
1227 |
-
essay_grades,
|
1228 |
-
general_comment,
|
1229 |
-
specific_comment,
|
1230 |
-
essay_year,
|
1231 |
-
reference,
|
1232 |
-
]
|
1233 |
-
rows.append(row)
|
1234 |
-
return rows
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|