zhangir-azerbayev commited on
Commit
4fa1d7a
1 Parent(s): 1f7a36e

ocw courses dataset

Browse files
Files changed (2) hide show
  1. ocw.py +154 -0
  2. test_dataloader.py +6 -0
ocw.py ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """OCWCourses"""
2
+
3
+
4
+ import json
5
+ import os
6
+ import re
7
+ import zipfile
8
+
9
+ import datasets
10
+
11
+ logger = datasets.logging.get_logger(__name__)
12
+
13
+ _CITATION = """\
14
+ @misc{lewkowycz2022solving,
15
+ title={Solving Quantitative Reasoning Problems with Language Models},
16
+ author={Aitor Lewkowycz and Anders Andreassen and David Dohan and Ethan Dyer and Henryk Michalewski and Vinay Ramasesh and Ambrose Slone and Cem Anil and Imanol Schlag and Theo Gutman-Solo and Yuhuai Wu and Behnam Neyshabur and Guy Gur-Ari and Vedant Misra},
17
+ year={2022},
18
+ eprint={2206.14858},
19
+ archivePrefix={arXiv},
20
+ primaryClass={cs.CL}
21
+ }
22
+ """
23
+
24
+ _DESCRIPTION = """\
25
+ OCWCourses
26
+ """
27
+ _URL = "https://openreview.net/forum?id=IFXTZERXdM7"
28
+ _URLS = {
29
+ "test": "https://openreview.net/attachment?id=IFXTZERXdM7&name=supplementary_material",
30
+ }
31
+
32
+ def _read_from_zip(zip_path, text_filename='minerva_supplement/ocw_courses_problems.text'):
33
+ with zipfile.ZipFile(zip_path, 'r') as myzip:
34
+ with myzip.open(text_filename) as myfile:
35
+ return myfile.read().decode('utf-8')
36
+
37
+
38
+ def _parse_latex(text):
39
+ problems = re.findall(r'\\textbf{Problem:}(.*?)\\textbf{Solution:}', text, re.DOTALL)
40
+ solutions = re.findall(r'\\textbf{Solution:}(.*?)(\\textbf{Problem:}|$)', text, re.DOTALL)
41
+
42
+ parsed_list = []
43
+ for problem, solution in zip(problems, solutions):
44
+ parsed_list.append({
45
+ 'problem': problem.strip(),
46
+ 'solution': solution[0].strip()
47
+ })
48
+
49
+ return parsed_list
50
+
51
+ def _last_boxed_only_string(string):
52
+
53
+ idx = string.rfind("\\boxed")
54
+ if "\\boxed " in string:
55
+ return "\\boxed " + string.split("\\boxed ")[-1].split("$")[0]
56
+ if idx < 0:
57
+ idx = string.rfind("\\fbox")
58
+ if idx < 0:
59
+ return None
60
+
61
+ i = idx
62
+ right_brace_idx = None
63
+ num_left_braces_open = 0
64
+ while i < len(string):
65
+ if string[i] == "{":
66
+ num_left_braces_open += 1
67
+ if string[i] == "}":
68
+ num_left_braces_open -= 1
69
+ if num_left_braces_open == 0:
70
+ right_brace_idx = i
71
+ break
72
+ i += 1
73
+
74
+ if right_brace_idx is None:
75
+ retval = None
76
+ else:
77
+ retval = string[idx : right_brace_idx + 1]
78
+
79
+ return retval
80
+
81
+ def _remove_boxed(s):
82
+ if "\\boxed " in s:
83
+ left = "\\boxed "
84
+ assert s[: len(left)] == left
85
+ return s[len(left) :]
86
+
87
+ left = "\\boxed{"
88
+
89
+ assert s[: len(left)] == left
90
+ assert s[-1] == "}"
91
+
92
+ return s[len(left) : -1]
93
+
94
+
95
+ class OCWConfig(datasets.BuilderConfig):
96
+ """BuilderConfig for OCW."""
97
+
98
+ def __init__(self, **kwargs):
99
+ """BuilderConfig for OCW.
100
+
101
+ Args:
102
+ **kwargs: keyword arguments forwarded to super.
103
+ """
104
+ super(OCWConfig, self).__init__(**kwargs)
105
+
106
+
107
+ class OCWCourses(datasets.GeneratorBasedBuilder):
108
+ """OCWCourses"""
109
+
110
+ BUILDER_CONFIGS = [
111
+ OCWConfig(
112
+ name="ocwcourses",
113
+ version=datasets.Version("1.0.0", ""),
114
+ description="OCWCourses",
115
+ ),
116
+ ]
117
+
118
+ def _info(self):
119
+ return datasets.DatasetInfo(
120
+ description=_DESCRIPTION,
121
+ features=datasets.Features(
122
+ {
123
+ "problem": datasets.Value("string"),
124
+ "solution": datasets.Value("string"),
125
+ "answer": datasets.Value("string"),
126
+ }
127
+ ),
128
+ # No default supervised_keys (as we have to pass both question
129
+ # and context as input).
130
+ supervised_keys=None,
131
+ homepage="https://openreview.net/forum?id=IFXTZERXdM7",
132
+ citation=_CITATION,
133
+ )
134
+
135
+ def _split_generators(self, dl_manager):
136
+ downloaded_files = dl_manager.download_and_extract(_URLS)
137
+
138
+ return [
139
+ datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["test"]}),
140
+ ]
141
+
142
+ def _generate_examples(self, filepath):
143
+ """This function returns the examples in the raw (text) form."""
144
+ logger.info("generating examples from = %s", filepath)
145
+
146
+ with open(os.path.join(filepath, "minerva_supplement/ocw_courses_problems.tex")) as f:
147
+ text = f.read()
148
+
149
+ parsed_problems_solutions = _parse_latex(text)
150
+
151
+ rows = [{**x, "answer": _remove_boxed(_last_boxed_only_string(x["solution"]))} for x in parsed_problems_solutions]
152
+
153
+ for key, row in enumerate(rows):
154
+ yield key, row
test_dataloader.py ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ from datasets import load_dataset
2
+
3
+ data = load_dataset("ocw.py")
4
+
5
+ [print(row) for row in data["test"]]
6
+ print(data)