Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,3 +1,232 @@
|
|
1 |
-
---
|
2 |
-
license: apache-2.0
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
task_categories:
|
4 |
+
- text-classification
|
5 |
+
language:
|
6 |
+
- en
|
7 |
+
tags:
|
8 |
+
- data-preprocessing
|
9 |
+
- automl
|
10 |
+
- quality-issues
|
11 |
+
- benchmarks
|
12 |
+
size_categories:
|
13 |
+
- 1K<n<10K
|
14 |
+
- 10K<n<100K
|
15 |
+
---
|
16 |
+
|
17 |
+
# Data Preprocessing AutoML Benchmarks
|
18 |
+
|
19 |
+
This repository contains text classification datasets with known data quality issues for preprocessing research in AutoML.
|
20 |
+
|
21 |
+
## Dataset Categories
|
22 |
+
|
23 |
+
### Redundancy Issues
|
24 |
+
- **ag_news**: News categorization with topic overlap
|
25 |
+
- **twenty_newsgroups**: Newsgroup posts with cross-posting
|
26 |
+
|
27 |
+
### Class Imbalance Issues
|
28 |
+
- **yelp_polarity**: Sentiment analysis with rating bias
|
29 |
+
- **sms_spam**: Spam detection with severe imbalance
|
30 |
+
|
31 |
+
### Label Noise Issues
|
32 |
+
- **imdb**: Movie reviews with subjective labels
|
33 |
+
- **amazon_polarity**: Product reviews with rating inconsistencies
|
34 |
+
|
35 |
+
### Outlier Issues
|
36 |
+
- **emotion**: Twitter emotion with length outliers
|
37 |
+
- **financial_phrasebank**: Financial sentiment with domain outliers
|
38 |
+
|
39 |
+
### Clean Baselines
|
40 |
+
- **trec**: Question classification with clean labels
|
41 |
+
|
42 |
+
## Dataset Structure
|
43 |
+
|
44 |
+
Each dataset contains:
|
45 |
+
- `train.csv`: Training split (~75% of original training data)
|
46 |
+
- `validation.csv`: Validation split (~25% of original training data)
|
47 |
+
- `test.csv`: Test split (original test set preserved)
|
48 |
+
|
49 |
+
All datasets have consistent columns:
|
50 |
+
- `text`: Input text
|
51 |
+
- `label`: Target label (integer encoded)
|
52 |
+
|
53 |
+
**Important**: Original test sets are preserved to maintain methodological integrity and enable comparison with published benchmarks.
|
54 |
+
|
55 |
+
## Usage
|
56 |
+
|
57 |
+
```python
|
58 |
+
from datasets import load_dataset
|
59 |
+
|
60 |
+
# Load a specific dataset
|
61 |
+
dataset = load_dataset("MothMalone/data-preprocessing-automl-benchmarks", "ag_news")
|
62 |
+
|
63 |
+
# Access splits
|
64 |
+
train_data = dataset["train"]
|
65 |
+
val_data = dataset["validation"]
|
66 |
+
test_data = dataset["test"]
|
67 |
+
```
|
68 |
+
|
69 |
+
## Metadata
|
70 |
+
|
71 |
+
ag_news:
|
72 |
+
class_names:
|
73 |
+
- World
|
74 |
+
- Sports
|
75 |
+
- Business
|
76 |
+
- Technology
|
77 |
+
description: News categorization with 4 classes, known for similar content across
|
78 |
+
categories
|
79 |
+
name: AG News Classification
|
80 |
+
num_classes: 4
|
81 |
+
original_test_samples: 7600
|
82 |
+
original_train_samples: 120000
|
83 |
+
quality_issues:
|
84 |
+
- redundancy
|
85 |
+
- similar_content
|
86 |
+
- topic_overlap
|
87 |
+
target_column: label
|
88 |
+
task_type: multi_classification
|
89 |
+
test_samples: 7600
|
90 |
+
text_columns:
|
91 |
+
- text
|
92 |
+
total_samples: 127600
|
93 |
+
train_samples: 90000
|
94 |
+
validation_samples: 30000
|
95 |
+
amazon_polarity:
|
96 |
+
class_names:
|
97 |
+
- negative
|
98 |
+
- positive
|
99 |
+
description: Amazon reviews with noisy sentiment labels
|
100 |
+
name: Amazon Product Reviews
|
101 |
+
num_classes: 2
|
102 |
+
original_test_samples: 400000
|
103 |
+
original_train_samples: 3600000
|
104 |
+
quality_issues:
|
105 |
+
- label_noise
|
106 |
+
- rating_inconsistency
|
107 |
+
target_column: label
|
108 |
+
task_type: binary_classification
|
109 |
+
test_samples: 400000
|
110 |
+
text_columns:
|
111 |
+
- text
|
112 |
+
total_samples: 4000000
|
113 |
+
train_samples: 2700000
|
114 |
+
validation_samples: 900000
|
115 |
+
emotion:
|
116 |
+
class_names:
|
117 |
+
- sadness
|
118 |
+
- joy
|
119 |
+
- love
|
120 |
+
- anger
|
121 |
+
- fear
|
122 |
+
- surprise
|
123 |
+
description: Twitter emotion classification with text length outliers
|
124 |
+
name: Emotion Classification
|
125 |
+
num_classes: 6
|
126 |
+
original_test_samples: 41681
|
127 |
+
original_train_samples: 333447
|
128 |
+
quality_issues:
|
129 |
+
- length_outliers
|
130 |
+
- text_anomalies
|
131 |
+
target_column: label
|
132 |
+
task_type: multi_classification
|
133 |
+
test_samples: 41681
|
134 |
+
text_columns:
|
135 |
+
- text
|
136 |
+
total_samples: 375128
|
137 |
+
train_samples: 250085
|
138 |
+
validation_samples: 83362
|
139 |
+
imdb:
|
140 |
+
class_names:
|
141 |
+
- negative
|
142 |
+
- positive
|
143 |
+
description: Movie reviews with subjective sentiment labels and borderline cases
|
144 |
+
name: IMDB Movie Reviews
|
145 |
+
num_classes: 2
|
146 |
+
original_test_samples: 25000
|
147 |
+
original_train_samples: 25000
|
148 |
+
quality_issues:
|
149 |
+
- label_noise
|
150 |
+
- subjective_labels
|
151 |
+
- borderline_cases
|
152 |
+
target_column: label
|
153 |
+
task_type: binary_classification
|
154 |
+
test_samples: 25000
|
155 |
+
text_columns:
|
156 |
+
- text
|
157 |
+
total_samples: 50000
|
158 |
+
train_samples: 18750
|
159 |
+
validation_samples: 6250
|
160 |
+
twenty_newsgroups:
|
161 |
+
class_names:
|
162 |
+
- alt.atheism
|
163 |
+
- comp.graphics
|
164 |
+
- comp.os.ms-windows.misc
|
165 |
+
- comp.sys.ibm.pc.hardware
|
166 |
+
- comp.sys.mac.hardware
|
167 |
+
- comp.windows.x
|
168 |
+
- misc.forsale
|
169 |
+
- rec.autos
|
170 |
+
- rec.motorcycles
|
171 |
+
- rec.sport.baseball
|
172 |
+
- rec.sport.hockey
|
173 |
+
- sci.crypt
|
174 |
+
- sci.electronics
|
175 |
+
- sci.med
|
176 |
+
- sci.space
|
177 |
+
- soc.religion.christian
|
178 |
+
- talk.politics.guns
|
179 |
+
- talk.politics.mideast
|
180 |
+
- talk.politics.misc
|
181 |
+
- talk.religion.misc
|
182 |
+
description: Newsgroup posts with overlapping topics and cross-posting
|
183 |
+
name: 20 Newsgroups
|
184 |
+
num_classes: 20
|
185 |
+
original_test_samples: 7532
|
186 |
+
original_train_samples: 11314
|
187 |
+
quality_issues:
|
188 |
+
- redundancy
|
189 |
+
- cross_posting
|
190 |
+
- similar_topics
|
191 |
+
target_column: label
|
192 |
+
task_type: multi_classification
|
193 |
+
test_samples: 7532
|
194 |
+
text_columns:
|
195 |
+
- text
|
196 |
+
total_samples: 18846
|
197 |
+
train_samples: 8485
|
198 |
+
validation_samples: 2829
|
199 |
+
yelp_polarity:
|
200 |
+
class_names:
|
201 |
+
- negative
|
202 |
+
- positive
|
203 |
+
description: Yelp reviews with positive/negative sentiment, naturally imbalanced
|
204 |
+
name: Yelp Review Polarity
|
205 |
+
num_classes: 2
|
206 |
+
original_test_samples: 38000
|
207 |
+
original_train_samples: 560000
|
208 |
+
quality_issues:
|
209 |
+
- moderate_imbalance
|
210 |
+
- rating_bias
|
211 |
+
target_column: label
|
212 |
+
task_type: binary_classification
|
213 |
+
test_samples: 38000
|
214 |
+
text_columns:
|
215 |
+
- text
|
216 |
+
total_samples: 598000
|
217 |
+
train_samples: 420000
|
218 |
+
validation_samples: 140000
|
219 |
+
|
220 |
+
|
221 |
+
## Citation
|
222 |
+
|
223 |
+
If you use these datasets in your research, please cite the original sources and this collection:
|
224 |
+
|
225 |
+
```bibtex
|
226 |
+
@misc{mothmalone2024preprocessing,
|
227 |
+
title={Data Preprocessing AutoML Benchmarks},
|
228 |
+
author={MothMalone},
|
229 |
+
year={2024},
|
230 |
+
url={https://huggingface.co/datasets/MothMalone/data-preprocessing-automl-benchmarks}
|
231 |
+
}
|
232 |
+
```
|