Update README.md
Browse files
README.md
CHANGED
@@ -38,7 +38,6 @@ task_ids:
|
|
38 |
- tabular-multi-label-classification
|
39 |
---
|
40 |
|
41 |
-
|
42 |
# Dataset Card for "AdapTable-full" - Dataset of Few-shot Tasks from Tables
|
43 |
|
44 |
## Table of Contents
|
@@ -74,39 +73,81 @@ task_ids:
|
|
74 |
|
75 |
### Dataset Summary
|
76 |
|
77 |
-
The AdapTable dataset consists of tables that naturally occur on the web
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
* [AdapTable-
|
86 |
-
|
87 |
-
*
|
88 |
-
* [AdapTable-
|
89 |
-
* [AdapTable-
|
90 |
-
* [AdapTable-
|
91 |
-
|
92 |
-
*
|
93 |
-
* [AdapTable-
|
94 |
-
* [AdapTable-
|
95 |
-
* [AdapTable-
|
96 |
-
* [AdapTable-
|
97 |
-
* [AdapTable-
|
98 |
-
* [AdapTable-
|
99 |
-
* [AdapTable-
|
100 |
-
* [AdapTable-
|
101 |
-
* [AdapTable-
|
102 |
-
*
|
103 |
-
* [AdapTable-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
104 |
|
105 |
### Supported Tasks and Leaderboards
|
106 |
|
107 |
-
Since the tables come from the web, the distribution of tasks and topics is very broad. The shape of our dataset is very wide, i.e., we have 1000's tasks, while each task has only a few examples, compared to most current NLP datasets which are very deep, i.e., 10s of tasks with many examples. This implies that our dataset covers a broad range of potential tasks, e.g., multiple-choice, question-answering, table-question-answering, text-classification, etc.
|
108 |
|
109 |
-
The intended use of this dataset is to improve few-shot performance by
|
110 |
|
111 |
### Languages
|
112 |
|
@@ -124,11 +165,11 @@ There are also additional meta-data fields such as 'pageTitle', 'title', 'output
|
|
124 |
|
125 |
'task': task identifier
|
126 |
|
127 |
-
'input': column elements of a specific row in table.
|
128 |
|
129 |
'options': for multiple choice classification, it provides the options to choose from.
|
130 |
|
131 |
-
'output': target column element of same row as input.
|
132 |
|
133 |
'pageTitle': the title of the page containing the table.
|
134 |
|
@@ -140,7 +181,7 @@ There are also additional meta-data fields such as 'pageTitle', 'title', 'output
|
|
140 |
|
141 |
### Data Splits
|
142 |
|
143 |
-
AdapTable
|
144 |
|
145 |
## Dataset Creation
|
146 |
|
@@ -149,13 +190,13 @@ AdapTable-full does not come with additional data splits.
|
|
149 |
How do we convert tables to few-shot tasks?
|
150 |
Unlike unstructured text, structured data in the form of tables lends itself easily to the few-shot task format. Given a table where each row is an instance of a similar class and the columns describe the attributes of each instance, we can turn each row into a task example to predict one attribute given the others. When the table has more than one row, we instantly have multiple examples of this task by using each row as a single example, and thus each table becomes a few-shot dataset for a particular task.
|
151 |
|
152 |
-
The few-shot setting in this
|
153 |
|
154 |
### Source Data
|
155 |
|
156 |
#### Initial Data Collection and Normalization
|
157 |
|
158 |
-
The data processing
|
159 |
|
160 |
#### Who are the source language producers?
|
161 |
|
@@ -166,10 +207,11 @@ The dataset is extracted from [WDC Web Table Corpora](http://webdatacommons.org/
|
|
166 |
#### Annotation process
|
167 |
|
168 |
No manual annotation process used.
|
|
|
169 |
|
170 |
#### Who are the annotators?
|
171 |
|
172 |
-
|
173 |
|
174 |
### Personal and Sensitive Information
|
175 |
|
@@ -181,12 +223,12 @@ The data was extracted from [WDC Web Table Corpora](http://webdatacommons.org/we
|
|
181 |
|
182 |
The purpose of this dataset is to help develop models that are better at few-shot learning and have higher few-shot performance by fine-tuning few-shot tasks extracted from tables.
|
183 |
|
184 |
-
While tables have a similar structure to few-shot tasks and we do see an improved performance on few-shot tasks in our paper, we want to make clear that
|
185 |
|
186 |
### Discussion of Biases
|
187 |
|
188 |
-
Since our dataset contains tables that are scraped from the web, it will also contain many toxic, racist, sexist, and otherwise harmful biases and texts. We have not run any analysis on the biases prevalent in our datasets. Neither have we explicitly filtered the content
|
189 |
-
This implies that a model trained on our dataset will reinforce harmful biases and toxic text that exist in our dataset.
|
190 |
|
191 |
### Other Known Limitations
|
192 |
|
|
|
38 |
- tabular-multi-label-classification
|
39 |
---
|
40 |
|
|
|
41 |
# Dataset Card for "AdapTable-full" - Dataset of Few-shot Tasks from Tables
|
42 |
|
43 |
## Table of Contents
|
|
|
73 |
|
74 |
### Dataset Summary
|
75 |
|
76 |
+
The AdapTable dataset consists of tables that naturally occur on the web and that are formatted as few-shot tasks for fine-tuning language models to improve their few-shot performance.
|
77 |
+
|
78 |
+
There are several dataset versions available:
|
79 |
+
|
80 |
+
* [AdapTable-full](https://huggingface.co/datasets/MicPie/adaptable_full): Starting from the initial WTC corpus of 50M tables, we apply our tables-to-tasks procedure to produce our resulting dataset, [AdapTable-full](https://huggingface.co/datasets/MicPie/adaptable_full), which comprises 413,350 tasks from 23,744 unique websites.
|
81 |
+
|
82 |
+
* [AdapTable-unique](https://huggingface.co/datasets/MicPie/adaptable_unique): This is the same as [AdapTable-full](https://huggingface.co/datasets/MicPie/adaptable_full) but filtered to have a maximum of one task per website. [AdapTable-unique](https://huggingface.co/datasets/MicPie/adaptable_unique) contains exactly 23,744 tasks from 23,744 websites.
|
83 |
+
|
84 |
+
* [AdapTable-5k](https://huggingface.co/datasets/MicPie/adaptable_5k): This dataset uses 5k random tables from the full dataset.
|
85 |
+
|
86 |
+
* AdapTable data subsets based on a manual human quality rating:
|
87 |
+
* [AdapTable-rated-low](https://huggingface.co/datasets/MicPie/adaptable_rated-low)
|
88 |
+
* [AdapTable-rated-medium](https://huggingface.co/datasets/MicPie/adaptable_rated-medium)
|
89 |
+
* [AdapTable-rated-high](https://huggingface.co/datasets/MicPie/adaptable_rated-high)
|
90 |
+
|
91 |
+
* AdapTable data subsets based on the website of origin:
|
92 |
+
* [AdapTable-baseball.fantasysports.yahoo.com](https://huggingface.co/datasets/MicPie/adaptable_baseball.fantasysports.yahoo.com)
|
93 |
+
* [AdapTable-bulbapedia.bulbagarden.net](https://huggingface.co/datasets/MicPie/adaptable_bulbapedia.bulbagarden.net)
|
94 |
+
* [AdapTable-cappex.com](https://huggingface.co/datasets/MicPie/adaptable_cappex.com)
|
95 |
+
* [AdapTable-cram.com](https://huggingface.co/datasets/MicPie/adaptable_cram.com)
|
96 |
+
* [AdapTable-dividend.com](https://huggingface.co/datasets/MicPie/adaptable_dividend.com)
|
97 |
+
* [AdapTable-dummies.com](https://huggingface.co/datasets/MicPie/adaptable_dummies.com)
|
98 |
+
* [AdapTable-en.wikipedia.org](https://huggingface.co/datasets/MicPie/adaptable_en.wikipedia.org)
|
99 |
+
* [AdapTable-ensembl.org](https://huggingface.co/datasets/MicPie/adaptable_ensembl.org)
|
100 |
+
* [AdapTable-gamefaqs.com](https://huggingface.co/datasets/MicPie/adaptable_gamefaqs.com)
|
101 |
+
* [AdapTable-mgoblog.com](https://huggingface.co/datasets/MicPie/adaptable_mgoblog.com)
|
102 |
+
* [AdapTable-mmo-champion.com](https://huggingface.co/datasets/MicPie/adaptable_mmo-champion.com)
|
103 |
+
* [AdapTable-msdn.microsoft.com](https://huggingface.co/datasets/MicPie/adaptable_msdn.microsoft.com)
|
104 |
+
* [AdapTable-phonearena.com](https://huggingface.co/datasets/MicPie/adaptable_phonearena.com)
|
105 |
+
* [AdapTable-sittercity.com](https://huggingface.co/datasets/MicPie/adaptable_sittercity.com)
|
106 |
+
* [AdapTable-sporcle.com](https://huggingface.co/datasets/MicPie/adaptable_sporcle.com)
|
107 |
+
* [AdapTable-studystack.com](https://huggingface.co/datasets/MicPie/adaptable_studystack.com)
|
108 |
+
* [AdapTable-support.google.com](https://huggingface.co/datasets/MicPie/adaptable_support.google.com)
|
109 |
+
* [AdapTable-w3.org](https://huggingface.co/datasets/MicPie/adaptable_w3.org)
|
110 |
+
* [AdapTable-wiki.openmoko.org](https://huggingface.co/datasets/MicPie/adaptable_wiki.openmoko.org)
|
111 |
+
* [AdapTable-wkdu.org](https://huggingface.co/datasets/MicPie/adaptable_wkdu.org)
|
112 |
+
|
113 |
+
* AdapTable data subsets based on clustering (for the clustering details please see our publication):
|
114 |
+
* [AdapTable-cluster00](https://huggingface.co/datasets/MicPie/adaptable_cluster00)
|
115 |
+
* [AdapTable-cluster01](https://huggingface.co/datasets/MicPie/adaptable_cluster01)
|
116 |
+
* [AdapTable-cluster02](https://huggingface.co/datasets/MicPie/adaptable_cluster02)
|
117 |
+
* [AdapTable-cluster03](https://huggingface.co/datasets/MicPie/adaptable_cluster03)
|
118 |
+
* [AdapTable-cluster04](https://huggingface.co/datasets/MicPie/adaptable_cluster04)
|
119 |
+
* [AdapTable-cluster05](https://huggingface.co/datasets/MicPie/adaptable_cluster05)
|
120 |
+
* [AdapTable-cluster06](https://huggingface.co/datasets/MicPie/adaptable_cluster06)
|
121 |
+
* [AdapTable-cluster07](https://huggingface.co/datasets/MicPie/adaptable_cluster07)
|
122 |
+
* [AdapTable-cluster08](https://huggingface.co/datasets/MicPie/adaptable_cluster08)
|
123 |
+
* [AdapTable-cluster09](https://huggingface.co/datasets/MicPie/adaptable_cluster09)
|
124 |
+
* [AdapTable-cluster10](https://huggingface.co/datasets/MicPie/adaptable_cluster10)
|
125 |
+
* [AdapTable-cluster11](https://huggingface.co/datasets/MicPie/adaptable_cluster11)
|
126 |
+
* [AdapTable-cluster12](https://huggingface.co/datasets/MicPie/adaptable_cluster12)
|
127 |
+
* [AdapTable-cluster13](https://huggingface.co/datasets/MicPie/adaptable_cluster13)
|
128 |
+
* [AdapTable-cluster14](https://huggingface.co/datasets/MicPie/adaptable_cluster14)
|
129 |
+
* [AdapTable-cluster15](https://huggingface.co/datasets/MicPie/adaptable_cluster15)
|
130 |
+
* [AdapTable-cluster16](https://huggingface.co/datasets/MicPie/adaptable_cluster16)
|
131 |
+
* [AdapTable-cluster17](https://huggingface.co/datasets/MicPie/adaptable_cluster17)
|
132 |
+
* [AdapTable-cluster18](https://huggingface.co/datasets/MicPie/adaptable_cluster18)
|
133 |
+
* [AdapTable-cluster19](https://huggingface.co/datasets/MicPie/adaptable_cluster19)
|
134 |
+
* [AdapTable-cluster20](https://huggingface.co/datasets/MicPie/adaptable_cluster20)
|
135 |
+
* [AdapTable-cluster21](https://huggingface.co/datasets/MicPie/adaptable_cluster21)
|
136 |
+
* [AdapTable-cluster22](https://huggingface.co/datasets/MicPie/adaptable_cluster22)
|
137 |
+
* [AdapTable-cluster23](https://huggingface.co/datasets/MicPie/adaptable_cluster23)
|
138 |
+
* [AdapTable-cluster24](https://huggingface.co/datasets/MicPie/adaptable_cluster24)
|
139 |
+
* [AdapTable-cluster25](https://huggingface.co/datasets/MicPie/adaptable_cluster25)
|
140 |
+
* [AdapTable-cluster26](https://huggingface.co/datasets/MicPie/adaptable_cluster26)
|
141 |
+
* [AdapTable-cluster27](https://huggingface.co/datasets/MicPie/adaptable_cluster27)
|
142 |
+
* [AdapTable-cluster28](https://huggingface.co/datasets/MicPie/adaptable_cluster28)
|
143 |
+
* [AdapTable-cluster29](https://huggingface.co/datasets/MicPie/adaptable_cluster29)
|
144 |
+
* [AdapTable-cluster-noise](https://huggingface.co/datasets/MicPie/adaptable_cluster-noise)
|
145 |
|
146 |
### Supported Tasks and Leaderboards
|
147 |
|
148 |
+
Since the tables come from the web, the distribution of tasks and topics is very broad. The shape of our dataset is very wide, i.e., we have 1000's of tasks, while each task has only a few examples, compared to most current NLP datasets which are very deep, i.e., 10s of tasks with many examples. This implies that our dataset covers a broad range of potential tasks, e.g., multiple-choice, question-answering, table-question-answering, text-classification, etc.
|
149 |
|
150 |
+
The intended use of this dataset is to improve few-shot performance by fine-tuning/pre-training on our dataset.
|
151 |
|
152 |
### Languages
|
153 |
|
|
|
165 |
|
166 |
'task': task identifier
|
167 |
|
168 |
+
'input': column elements of a specific row in the table.
|
169 |
|
170 |
'options': for multiple choice classification, it provides the options to choose from.
|
171 |
|
172 |
+
'output': target column element of the same row as input.
|
173 |
|
174 |
'pageTitle': the title of the page containing the table.
|
175 |
|
|
|
181 |
|
182 |
### Data Splits
|
183 |
|
184 |
+
The AdapTable datasets do not come with additional data splits.
|
185 |
|
186 |
## Dataset Creation
|
187 |
|
|
|
190 |
How do we convert tables to few-shot tasks?
|
191 |
Unlike unstructured text, structured data in the form of tables lends itself easily to the few-shot task format. Given a table where each row is an instance of a similar class and the columns describe the attributes of each instance, we can turn each row into a task example to predict one attribute given the others. When the table has more than one row, we instantly have multiple examples of this task by using each row as a single example, and thus each table becomes a few-shot dataset for a particular task.
|
192 |
|
193 |
+
The few-shot setting in this setup is significant: Tables often do not come with clear instructions for each field, so tasks may be underspecified if prompted in a zero-shot manner, but the intended task becomes clearer when examples are provided. This makes a good two-way match: The few-shot format is a perfect setup for table learning, and tables provide a natural dataset for few-shot training.
|
194 |
|
195 |
### Source Data
|
196 |
|
197 |
#### Initial Data Collection and Normalization
|
198 |
|
199 |
+
The data processing pipeline is explained in detail in our publication.
|
200 |
|
201 |
#### Who are the source language producers?
|
202 |
|
|
|
207 |
#### Annotation process
|
208 |
|
209 |
No manual annotation process used.
|
210 |
+
Only for the [AdapTable-rated-low](https://huggingface.co/datasets/MicPie/adaptable_rated-low), [AdapTable-rated-medium](https://huggingface.co/datasets/MicPie/adaptable_rated-medium), and [AdapTable-rated-high](https://huggingface.co/datasets/MicPie/adaptable_rated-high) manual annotations were carried out.
|
211 |
|
212 |
#### Who are the annotators?
|
213 |
|
214 |
+
People involved in the publication.
|
215 |
|
216 |
### Personal and Sensitive Information
|
217 |
|
|
|
223 |
|
224 |
The purpose of this dataset is to help develop models that are better at few-shot learning and have higher few-shot performance by fine-tuning few-shot tasks extracted from tables.
|
225 |
|
226 |
+
While tables have a similar structure to few-shot tasks and we do see an improved performance on few-shot tasks in our paper, we want to make clear that fine-tuning on tables also has its risks. First of all, since the tables are extracted from the web, they may contain user identities or otherwise sensitive information which a model might reveal at inference, or which could influence the learning process of a model in a negative way. Second, since tables are very diverse in nature, the model also trains on low-quality data or data with an unusual structure. While it is interesting that training on such data improves few-shot performance on downstream tasks, this could also imply that the model learns concepts that are very dissimilar to human concepts that would be useful for a certain downstream task. In other words, it is possible that the model learns weird things that are helpful on the evaluated downstream tasks, but might lead to bad out-of-distribution behavior.
|
227 |
|
228 |
### Discussion of Biases
|
229 |
|
230 |
+
Since our dataset contains tables that are scraped from the web, it will also contain many toxic, racist, sexist, and otherwise harmful biases and texts. We have not run any analysis on the biases prevalent in our datasets. Neither have we explicitly filtered the content.
|
231 |
+
This implies that a model trained on our dataset will potentially reinforce harmful biases and toxic text that exist in our dataset.
|
232 |
|
233 |
### Other Known Limitations
|
234 |
|