Datasets:
Tasks:
Text Generation
Modalities:
Text
Sub-tasks:
language-modeling
Languages:
English
Size:
100K - 1M
License:
Zhangir Azerbayev
commited on
Commit
·
cda48cc
1
Parent(s):
231bd79
updated readme
Browse files
README.md
CHANGED
@@ -56,7 +56,7 @@ This dataset contains only the training set of the [MATH dataset](https://github
|
|
56 |
# Data Preprocessing
|
57 |
This section describes any significant filtering and transformations made to various subsets of the data.
|
58 |
|
59 |
-
|
60 |
The arXiv.math dataset is large, heterogeneous, and contains a great deal of noise. We used the following heuristics
|
61 |
when choosing which files from arXiv.math source folders to include in the dataset:
|
62 |
- Keep only files with a `.tex` extension.
|
@@ -67,7 +67,9 @@ when choosing which files from arXiv.math source folders to include in the datas
|
|
67 |
- Include only articles in English, as determined by the [langdetect library](https://pypi.org/project/langdetect/). \n",
|
68 |
"\n",
|
69 |
- Exclude files shorter than 280 characters (characters counted after substring removal described below).
|
|
|
70 |
In addition, we apply the following transformations to arXiv.math texts:
|
|
|
71 |
- Delete everything outside of `\begin{document}` and `\end{document}`.
|
72 |
- Delete everything including or after `\Refs`, `\begin{thebibliography}`, or `\begin{bibdiv}`
|
73 |
- Delete comments.
|
@@ -75,7 +77,7 @@ In addition, we apply the following transformations to arXiv.math texts:
|
|
75 |
In [this notebook](https://github.com/zhangir-azerbayev/proof-pile/blob/main/analysis/arxiv_noisedetection.ipynb), we provide an analysis of the prevalence of noisy documents in the arXiv.math subset of the
|
76 |
proof-pile.
|
77 |
|
78 |
-
|
79 |
We only include questions that have at least 5 upvotes and an answer. We format Stack Exchange posts as follows
|
80 |
```
|
81 |
QUESTION [{num_upvotes} upvotes]: {text of question}
|
@@ -89,7 +91,7 @@ REPLY [{num_upvotes} votes]: {text of reply}
|
|
89 |
.
|
90 |
```
|
91 |
|
92 |
-
|
93 |
We converted `set.mm` into human-readable form by following the instructions in the [mm-extract repo](https://github.com/zhangir-azerbayev/mm-extract)
|
94 |
|
95 |
## Contributions
|
|
|
56 |
# Data Preprocessing
|
57 |
This section describes any significant filtering and transformations made to various subsets of the data.
|
58 |
|
59 |
+
**arXiv.math**
|
60 |
The arXiv.math dataset is large, heterogeneous, and contains a great deal of noise. We used the following heuristics
|
61 |
when choosing which files from arXiv.math source folders to include in the dataset:
|
62 |
- Keep only files with a `.tex` extension.
|
|
|
67 |
- Include only articles in English, as determined by the [langdetect library](https://pypi.org/project/langdetect/). \n",
|
68 |
"\n",
|
69 |
- Exclude files shorter than 280 characters (characters counted after substring removal described below).
|
70 |
+
|
71 |
In addition, we apply the following transformations to arXiv.math texts:
|
72 |
+
|
73 |
- Delete everything outside of `\begin{document}` and `\end{document}`.
|
74 |
- Delete everything including or after `\Refs`, `\begin{thebibliography}`, or `\begin{bibdiv}`
|
75 |
- Delete comments.
|
|
|
77 |
In [this notebook](https://github.com/zhangir-azerbayev/proof-pile/blob/main/analysis/arxiv_noisedetection.ipynb), we provide an analysis of the prevalence of noisy documents in the arXiv.math subset of the
|
78 |
proof-pile.
|
79 |
|
80 |
+
**Stack Exchange**
|
81 |
We only include questions that have at least 5 upvotes and an answer. We format Stack Exchange posts as follows
|
82 |
```
|
83 |
QUESTION [{num_upvotes} upvotes]: {text of question}
|
|
|
91 |
.
|
92 |
```
|
93 |
|
94 |
+
**set.mm**
|
95 |
We converted `set.mm` into human-readable form by following the instructions in the [mm-extract repo](https://github.com/zhangir-azerbayev/mm-extract)
|
96 |
|
97 |
## Contributions
|
utils.py
DELETED
@@ -1,53 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
import tarfile
|
3 |
-
from itertools import cycle
|
4 |
-
from shutil import get_terminal_size
|
5 |
-
from threading import Thread
|
6 |
-
from time import sleep
|
7 |
-
|
8 |
-
def make_archive(path):
|
9 |
-
with tarfile.open(path + ".tar.gz", "w:gz") as tar:
|
10 |
-
tar.add(path, arcname=os.path.sep)
|
11 |
-
os.system(f"rm -r {path}")
|
12 |
-
|
13 |
-
class Loader:
|
14 |
-
def __init__(self, desc="Loading...", end="Done!", timeout=0.1):
|
15 |
-
"""
|
16 |
-
A loader-like context manager
|
17 |
-
|
18 |
-
Args:
|
19 |
-
desc (str, optional): The loader's description. Defaults to "Loading...".
|
20 |
-
end (str, optional): Final print. Defaults to "Done!".
|
21 |
-
timeout (float, optional): Sleep time between prints. Defaults to 0.1.
|
22 |
-
"""
|
23 |
-
self.desc = desc
|
24 |
-
self.end = end
|
25 |
-
self.timeout = timeout
|
26 |
-
|
27 |
-
self._thread = Thread(target=self._animate, daemon=True)
|
28 |
-
self.steps = ["⢿", "⣻", "⣽", "⣾", "⣷", "⣯", "⣟", "⡿"]
|
29 |
-
self.done = False
|
30 |
-
|
31 |
-
def start(self):
|
32 |
-
self._thread.start()
|
33 |
-
return self
|
34 |
-
|
35 |
-
def _animate(self):
|
36 |
-
for c in cycle(self.steps):
|
37 |
-
if self.done:
|
38 |
-
break
|
39 |
-
print(f"\r{self.desc} {c}", flush=True, end="")
|
40 |
-
sleep(self.timeout)
|
41 |
-
|
42 |
-
def __enter__(self):
|
43 |
-
self.start()
|
44 |
-
|
45 |
-
def stop(self):
|
46 |
-
self.done = True
|
47 |
-
cols = get_terminal_size((80, 20)).columns
|
48 |
-
print("\r" + " " * cols, end="", flush=True)
|
49 |
-
print(f"\r{self.end}", flush=True)
|
50 |
-
|
51 |
-
def __exit__(self, exc_type, exc_value, tb):
|
52 |
-
# handle exceptions with those variables ^
|
53 |
-
self.stop()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|