Loading script

BLUEX_without_images_disabled.py

@@ -0,0 +1,115 @@
+"""OAB Exams dataset"""
+
+import datasets
+import pandas as pd
+import re
+from collections import defaultdict
+import os
+import json
+
+_CITATION = """@misc{almeida2023bluex,
+      title={BLUEX: A benchmark based on Brazilian Leading Universities Entrance eXams}, 
+      author={Thales Sales Almeida and Thiago Laitz and Giovana K. Bonás and Rodrigo Nogueira},
+      year={2023},
+      eprint={2307.05410},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL}
+}
+"""
+
+_DESCRIPTION = """
+Despite Portuguese being the fifth most widely spoken language, there is a lack of freely available resources for evaluating language models in Portuguese. This repository contains a multimodal dataset consisting of the two leading university entrance exams conducted in Brazil: Convest (Unicamp) and Fuvest (USP), spanning from 2018 to 2024. The dataset comprises a total of 1260 questions, of which 724 do not have accompanying images.
+"""
+
+_HOMEPAGE="https://github.com/Portuguese-Benchmark-Datasets/BLUEX"
+
+_URL = "portuguese-benchmark-datasets/BLUEX"
+_URL = "https://raw.githubusercontent.com/Portuguese-Benchmark-Datasets/BLUEX/main/data/bluex_dataset.zip"
+
+class BLUEX_without_images(datasets.GeneratorBasedBuilder):
+
+    VERSION = datasets.Version("1.1.0")
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "id": datasets.Value("string"),
+                    "question_number": datasets.Value("int32"),
+                    "exam_id": datasets.Value("string"),
+                    "exam_year": datasets.Value("string"),
+                    "university": datasets.Value("string"),
+                    "question_type": datasets.Sequence(datasets.Value("string")),
+                    "nullified": datasets.Value("bool"),
+                    "question": datasets.Value("string"),
+                    "choices": datasets.Sequence(feature={
+                        "text": datasets.Value("string"),
+                        "label": datasets.Value("string")
+                    }),
+                    "answerKey": datasets.Value("string"),
+                    }),
+            supervised_keys=None,
+            homepage=_HOMEPAGE,
+            citation=_CITATION,
+        )
+
+    def _split_generators(self, dl_manager):
+        #dataset = datasets.load_dataset(_URL, split="questions")
+        #remove questions that require images
+        #dataset = dataset.filter(lambda example: not example['IU'] and example['alternatives_type'] == 'string')
+        filedir = dl_manager.download_and_extract(_URL)
+        return [
+            datasets.SplitGenerator(
+                name=datasets.Split.TRAIN,
+                gen_kwargs={
+                    "filedir": os.path.join(filedir, 'questions')
+                }
+            )
+        ]
+
+    def _generate_examples(self, filedir):
+
+        for university in os.listdir(filedir):
+            years = sorted(os.listdir(os.path.join(filedir, university)))
+            for year in years:
+                days = [d for d in os.listdir(os.path.join(filedir, university, year)) if os.path.isdir(os.path.join(filedir, university, year, d))]
+                if len(days) == 0:
+                    days = ['']
+                days = sorted(days)
+                for day in days:
+                    if day == '':
+                        path = os.path.join(filedir, university, year)
+                    else:
+                        path = os.path.join(filedir, university, year, day)
+
+                    exam_id = f"{university}_{year}" if day == '' else f"{university}_{year}_{day.replace('day', '')}"
+                    filenames = sorted(os.listdir(path), key=lambda x: int(re.findall(r'\d+', x)[0]))
+                    for filename in filenames:
+                        if filename.endswith('.json'):
+                            with open(os.path.join(path, filename), 'r') as f:
+                                example = json.load(f)
+
+                            if example['IU'] or example['alternatives_type'] != 'string' or example['has_associated_images']:
+                                continue
+
+                            choices = {
+                                "text": [],
+                                "label": ["A", "B", "C", "D", "E"]
+                            }
+                            for alternative in example['alternatives']:
+                                choices['text'].append(alternative[3:].strip())
+                            choices['label'] = choices['label'][:len(choices['text'])]
+
+                            doc_id = f"{exam_id}_{example['number']}"
+                            yield doc_id, {
+                                "id": doc_id, 
+                                "question_number": example['number'], 
+                                "exam_id": exam_id, 
+                                "exam_year": year,
+                                "university": university,
+                                "question_type": example['subject'], 
+                                "nullified": None, 
+                                "question": example['question'], 
+                                "choices": choices,
+                                "answerKey": example['answer'] 
+                            }