- תיאור :
מערך הנתונים מכיל צמדים של שאלה-טבלה, והתשובה המתאימה. השאלות דורשות חשיבה רב-שלבית ופעולות נתונים שונות כגון השוואה, צבירה וחישוב אריתמטי. הטבלאות נבחרו באקראי בין טבלאות ויקיפדיה עם לפחות 8 שורות ו-5 עמודות.
(לפי הערות השימוש בתיעוד)
Dev: דיוק ממוצע על פני שלושה (לא חמישה) פיצולים של נתוני האימון. במילים אחרות, אימון על 'פיצול-{1,2,3}-רכב' ובדוק על 'פיצול-{1,2,3}-מפתח', בהתאמה, ואז ממוצע הדיוק.
מבחן: מתאמן על 'רכבת' ומבחן על 'מבחן'.
תיעוד נוסף : חקור על ניירות עם קוד
דף הבית : https://ppasupat.github.io/WikiTableQuestions/#usage-notes
קוד מקור :
tfds.structured.wiki_table_questions.WikiTableQuestionsגרסאות :
-
1.0.0(ברירת מחדל): שחרור ראשוני.
-
גודל הורדה :
65.36 MiBגודל ערכת נתונים:
237.24 MiBשמור אוטומטי במטמון ( תיעוד ): כן
פיצולים :
| לְפַצֵל | דוגמאות |
|---|---|
'split-1-dev' | 2,810 |
'split-1-train' | 11,321 |
'split-2-dev' | 2,838 |
'split-2-train' | 11,312 |
'split-3-dev' | 2,838 |
'split-3-train' | 11,311 |
'test' | 4,344 |
'train' | 14,149 |
- מבנה תכונה :
FeaturesDict({
'input_text': FeaturesDict({
'context': string,
'table': Sequence({
'column_header': string,
'content': string,
'row_number': int16,
}),
}),
'target_text': string,
})
- תיעוד תכונה :
| תכונה | מעמד | צוּרָה | Dtype | תיאור |
|---|---|---|---|---|
| FeaturesDict | ||||
| הקלד טקסט | FeaturesDict | |||
| input_text/context | מוֹתֵחַ | חוּט | ||
| input_text/טבלה | סדר פעולות | |||
| input_text/table/column_header | מוֹתֵחַ | חוּט | ||
| input_text/טבלה/תוכן | מוֹתֵחַ | חוּט | ||
| input_text/table/row_number | מוֹתֵחַ | int16 | ||
| target_text | מוֹתֵחַ | חוּט |
מפתחות בפיקוח (ראה
as_supervised):('input_text', 'target_text')איור ( tfds.show_examples ): לא נתמך.
דוגמאות ( tfds.as_dataframe ):
- ציטוט :
@inproceedings{pasupat-liang-2015-compositional,
title = "Compositional Semantic Parsing on Semi-Structured Tables",
author = "Pasupat, Panupong and
Liang, Percy",
booktitle = "Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)",
month = jul,
year = "2015",
address = "Beijing, China",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/P15-1142",
doi = "10.3115/v1/P15-1142",
pages = "1470--1480",
}