Add 21 feature selection benchmark datasets and dataset card
Browse files- .gitattributes +20 -0
- README.md +110 -0
- cao.hdf +3 -0
- coil-20.hdf +3 -0
- credit_default.hdf +3 -0
- german_credit.hdf +3 -0
- han.hdf +3 -0
- ionosphere.hdf +3 -0
- jannis.hdf +3 -0
- megawatt1.hdf +0 -0
- mice_protein.hdf +3 -0
- mnist.hdf +3 -0
- openml_586.hdf +3 -0
- openml_589.hdf +3 -0
- openml_607.hdf +3 -0
- openml_616.hdf +3 -0
- openml_618.hdf +3 -0
- openml_620.hdf +3 -0
- openml_637.hdf +3 -0
- otto.hdf +3 -0
- spam_base.hdf +3 -0
- spectf.hdf +3 -0
- svmguide3.hdf +3 -0
.gitattributes
CHANGED
|
@@ -58,3 +58,23 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 58 |
# Video files - compressed
|
| 59 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 60 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
# Video files - compressed
|
| 59 |
*.mp4 filter=lfs diff=lfs merge=lfs -text
|
| 60 |
*.webm filter=lfs diff=lfs merge=lfs -text
|
| 61 |
+
cao.hdf filter=lfs diff=lfs merge=lfs -text
|
| 62 |
+
coil-20.hdf filter=lfs diff=lfs merge=lfs -text
|
| 63 |
+
credit_default.hdf filter=lfs diff=lfs merge=lfs -text
|
| 64 |
+
german_credit.hdf filter=lfs diff=lfs merge=lfs -text
|
| 65 |
+
han.hdf filter=lfs diff=lfs merge=lfs -text
|
| 66 |
+
ionosphere.hdf filter=lfs diff=lfs merge=lfs -text
|
| 67 |
+
jannis.hdf filter=lfs diff=lfs merge=lfs -text
|
| 68 |
+
mice_protein.hdf filter=lfs diff=lfs merge=lfs -text
|
| 69 |
+
mnist.hdf filter=lfs diff=lfs merge=lfs -text
|
| 70 |
+
openml_586.hdf filter=lfs diff=lfs merge=lfs -text
|
| 71 |
+
openml_589.hdf filter=lfs diff=lfs merge=lfs -text
|
| 72 |
+
openml_607.hdf filter=lfs diff=lfs merge=lfs -text
|
| 73 |
+
openml_616.hdf filter=lfs diff=lfs merge=lfs -text
|
| 74 |
+
openml_618.hdf filter=lfs diff=lfs merge=lfs -text
|
| 75 |
+
openml_620.hdf filter=lfs diff=lfs merge=lfs -text
|
| 76 |
+
openml_637.hdf filter=lfs diff=lfs merge=lfs -text
|
| 77 |
+
otto.hdf filter=lfs diff=lfs merge=lfs -text
|
| 78 |
+
spam_base.hdf filter=lfs diff=lfs merge=lfs -text
|
| 79 |
+
spectf.hdf filter=lfs diff=lfs merge=lfs -text
|
| 80 |
+
svmguide3.hdf filter=lfs diff=lfs merge=lfs -text
|
README.md
ADDED
|
@@ -0,0 +1,110 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: Feature Selection Benchmark Datasets (HRLFS)
|
| 3 |
+
task_categories:
|
| 4 |
+
- tabular-classification
|
| 5 |
+
- tabular-regression
|
| 6 |
+
tags:
|
| 7 |
+
- feature-selection
|
| 8 |
+
- tabular
|
| 9 |
+
- reinforcement-learning
|
| 10 |
+
- benchmark
|
| 11 |
+
size_categories:
|
| 12 |
+
- 100K<n<1M
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# Feature Selection Benchmark Datasets (HRLFS)
|
| 16 |
+
|
| 17 |
+
This repository hosts the **21 benchmark datasets** used in our ACM TKDD paper:
|
| 18 |
+
|
| 19 |
+
> **Comprehend, Divide, and Conquer: Feature Subspace Exploration via Multi-Agent Hierarchical Reinforcement Learning**
|
| 20 |
+
> Weiliang Zhang, Xiaohan Huang, Yi Du, Ziyue Qiao, Qingqing Long, Zhen Meng, Yuanchun Zhou, Meng Xiao
|
| 21 |
+
> *ACM Transactions on Knowledge Discovery from Data (TKDD), 2026*
|
| 22 |
+
|
| 23 |
+
- 📄 Paper code: [https://github.com/coco11563/HARLFS](https://github.com/coco11563/HARLFS)
|
| 24 |
+
|
| 25 |
+
## Dataset Summary
|
| 26 |
+
|
| 27 |
+
The 21 datasets are publicly available and collected from the Feature Selection Benchmark, NCBI Gene Expression Omnibus (GEO), UCI, Kaggle, OpenML, libSVM, etc. They cover three task types — binary classification (C), multi-class classification (MC), and regression (R) — and span diverse fields including biology, finance, image, and synthetic data. Sample sizes range from 253 to 83,733 and feature dimensions from 21 to 20,670.
|
| 28 |
+
|
| 29 |
+
| Dataset | Task | #Samples | #Features | File |
|
| 30 |
+
|---|---|---|---|---|
|
| 31 |
+
| SpectF | C | 267 | 44 | `spectf.hdf` |
|
| 32 |
+
| SVMGuide3 | C | 1,243 | 21 | `svmguide3.hdf` |
|
| 33 |
+
| German Credit | C | 1,001 | 24 | `german_credit.hdf` |
|
| 34 |
+
| Credit Default | C | 30,000 | 25 | `credit_default.hdf` |
|
| 35 |
+
| SpamBase | C | 4,601 | 57 | `spam_base.hdf` |
|
| 36 |
+
| Megawatt1 | C | 253 | 38 | `megawatt1.hdf` |
|
| 37 |
+
| Ionosphere | C | 351 | 34 | `ionosphere.hdf` |
|
| 38 |
+
| Mice-Protein | MC | 1,080 | 77 | `mice_protein.hdf` |
|
| 39 |
+
| Coil-20 | MC | 1,440 | 400 | `coil-20.hdf` |
|
| 40 |
+
| MNIST | MC | 10,000 | 784 | `mnist.hdf` |
|
| 41 |
+
| Otto | MC | 61,878 | 93 | `otto.hdf` |
|
| 42 |
+
| Jannis | MC | 83,733 | 54 | `jannis.hdf` |
|
| 43 |
+
| Cao | MC | 4,186 | 13,488 | `cao.hdf` |
|
| 44 |
+
| Han | MC | 2,746 | 20,670 | `han.hdf` |
|
| 45 |
+
| Openml_586 | R | 1,000 | 25 | `openml_586.hdf` |
|
| 46 |
+
| Openml_589 | R | 1,000 | 25 | `openml_589.hdf` |
|
| 47 |
+
| Openml_607 | R | 1,000 | 50 | `openml_607.hdf` |
|
| 48 |
+
| Openml_616 | R | 500 | 50 | `openml_616.hdf` |
|
| 49 |
+
| Openml_618 | R | 1,000 | 50 | `openml_618.hdf` |
|
| 50 |
+
| Openml_620 | R | 1,000 | 25 | `openml_620.hdf` |
|
| 51 |
+
| Openml_637 | R | 500 | 50 | `openml_637.hdf` |
|
| 52 |
+
|
| 53 |
+
## Data Format
|
| 54 |
+
|
| 55 |
+
Each dataset is stored as a single HDF5 file with two keys:
|
| 56 |
+
|
| 57 |
+
- `raw_train` — training split (80% of samples)
|
| 58 |
+
- `raw_test` — test split (20% of samples)
|
| 59 |
+
|
| 60 |
+
Each key stores a `pandas.DataFrame` where **the last column is the label** and all preceding columns are features.
|
| 61 |
+
|
| 62 |
+
## Usage
|
| 63 |
+
|
| 64 |
+
Download the files with `huggingface_hub` and load them with `pandas`:
|
| 65 |
+
|
| 66 |
+
```python
|
| 67 |
+
import pandas as pd
|
| 68 |
+
from huggingface_hub import hf_hub_download
|
| 69 |
+
|
| 70 |
+
path = hf_hub_download(
|
| 71 |
+
repo_id="Shaow/Feature_Selection_Dataset",
|
| 72 |
+
filename="spam_base.hdf",
|
| 73 |
+
repo_type="dataset",
|
| 74 |
+
)
|
| 75 |
+
|
| 76 |
+
train = pd.read_hdf(path, key="raw_train")
|
| 77 |
+
test = pd.read_hdf(path, key="raw_test")
|
| 78 |
+
|
| 79 |
+
X_train, y_train = train.iloc[:, :-1].to_numpy(), train.iloc[:, -1].to_numpy()
|
| 80 |
+
X_test, y_test = test.iloc[:, :-1].to_numpy(), test.iloc[:, -1].to_numpy()
|
| 81 |
+
```
|
| 82 |
+
|
| 83 |
+
To reproduce the experiments in the paper, place the `.hdf` files under the `./data` directory of the [HARLFS repository](https://github.com/coco11563/HARLFS) and run, e.g.:
|
| 84 |
+
|
| 85 |
+
```bash
|
| 86 |
+
python HRLFS.py --dataset spam_base
|
| 87 |
+
```
|
| 88 |
+
|
| 89 |
+
Note: reading these files requires `pandas` and `tables` (PyTables): `pip install pandas tables`.
|
| 90 |
+
|
| 91 |
+
## Licensing
|
| 92 |
+
|
| 93 |
+
All datasets are redistributed from publicly available sources (Feature Selection Benchmark, NCBI GEO, UCI, Kaggle, OpenML, libSVM). Please refer to the original sources for their respective license terms; this collection is provided for research purposes only.
|
| 94 |
+
|
| 95 |
+
## Citation
|
| 96 |
+
|
| 97 |
+
If you use these datasets, please cite our paper:
|
| 98 |
+
|
| 99 |
+
```bibtex
|
| 100 |
+
@article{zhang2026comprehend,
|
| 101 |
+
title={Comprehend, Divide, and Conquer: Feature Subspace Exploration via Multi-Agent Hierarchical Reinforcement Learning},
|
| 102 |
+
author={Zhang, Weiliang and Huang, Xiaohan and Du, Yi and Qiao, Ziyue and Long, Qingqing and Meng, Zhen and Zhou, Yuanchun and Xiao, Meng},
|
| 103 |
+
journal={ACM Transactions on Knowledge Discovery from Data},
|
| 104 |
+
year={2026}
|
| 105 |
+
}
|
| 106 |
+
```
|
| 107 |
+
|
| 108 |
+
## Contact
|
| 109 |
+
|
| 110 |
+
For questions, please contact the corresponding author: **Meng Xiao** (shaow@cnic.cn).
|
cao.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:054b52e77de71e6f7ed35db991ebaaa1a2b1e0e8252a1937ae06165390bbbc1e
|
| 3 |
+
size 452048632
|
coil-20.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b130e7f7b235c20b457c6c1b6c0b610f1a84b694f36c706257c9d30874644014
|
| 3 |
+
size 4653160
|
credit_default.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:92b3527892ae8e578310deccc54067021eac2df4421c33057532ae75b01c08a8
|
| 3 |
+
size 6256000
|
german_credit.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d0cf6fe1c4afcf6676f159b29d9b2af88c5ef781e78cf09ea25897186c844661
|
| 3 |
+
size 219784
|
han.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:eb84a1138df42c222192c43d4e0899972cf506e9d18285d3e7053766688b052f
|
| 3 |
+
size 454601896
|
ionosphere.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8ccdfb0f8bed00114b239d2f42d223d5f8afba2ec65be6cfaeacfb1c031fdbe9
|
| 3 |
+
size 116864
|
jannis.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c3d73f37a04e56317140e2bc724976795af685609c6c748f40c339066abcd9b7
|
| 3 |
+
size 47038616
|
megawatt1.hdf
ADDED
|
Binary file (90.6 kB). View file
|
|
|
mice_protein.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7c65c53a554b3c94b68b3adfe33f99db9f590fa943e41afcf507cbf2bec2586d
|
| 3 |
+
size 695944
|
mnist.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:76b213f8118c0b2cf2318c9a7e9cd77428ebbaf0d19fe20e706ff22d1cd9878d
|
| 3 |
+
size 62914408
|
openml_586.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:75f3117b44fc929c9906fbda3b1b2255b6ffdc3c71f3aabe4ca71d23bb1c0f26
|
| 3 |
+
size 227784
|
openml_589.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6e3cd58c99c6d04d6c049b807cfa834124d15f1c6a89f358c826a6df109a106d
|
| 3 |
+
size 227784
|
openml_607.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5e81d62e3462278911ac62ab45c92aa3b968d02c79c6be62e7dc9b9376525623
|
| 3 |
+
size 428192
|
openml_616.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8a35ca146d639bd5799847f414d17fd030f0dfc20ebb09c44c798601c4d60422
|
| 3 |
+
size 220584
|
openml_618.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:fae216811ea4370c28fa88583a0d53d8f46491d57dd87333fc68cfb659bd0fef
|
| 3 |
+
size 428192
|
openml_620.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:711b17d9c9ba21ca6012126750e6ed6df5f94e551cac9386a081bccfbd505087
|
| 3 |
+
size 227784
|
openml_637.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8d8bfcb386091cbadafe557d624edec7ae369189658984ad6be91a70f7e5045b
|
| 3 |
+
size 220584
|
otto.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c464cd41f7b0ffe35b7ec47c5e389315548e62c172933e8b245776bcbf65fe2f
|
| 3 |
+
size 47036568
|
spam_base.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ab8f685da14a94439d17355d50938c19a2441cf53af91aab75be19234ce40920
|
| 3 |
+
size 2187672
|
spectf.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0790c194a54bb2a959d3d2e8718984ae5123a836f6720a8266dd29ffcc424d6e
|
| 3 |
+
size 111208
|
svmguide3.hdf
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7434738df0cf4a013e9f6c32db5c1ce6636f368478e314eaf8e661a0c849b454
|
| 3 |
+
size 242096
|