ivopersus commited on
Commit
16e7710
·
verified ·
1 Parent(s): 3f7efd2

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +29 -50
README.md CHANGED
@@ -1,84 +1,63 @@
1
  ---
2
- license: cc-by-4.0
3
  language:
4
  - en
5
- pretty_name: UCI Tabular Benchmark Sample
6
  task_categories:
7
  - tabular-classification
8
  - tabular-regression
9
  tags:
10
- - uci
11
  - tabular
12
  - classification
13
  - binary
14
- - multiclass
15
  - regression
 
 
 
 
 
16
  size_categories:
17
  - 10K<n<100K
18
- - 100K<n<1M
19
  ---
20
 
21
- # UCI Tabular Benchmark Sample
22
- This repository contains a small collection of tabular datasets mirrored from the UCI Machine Learning Repository and prepared for convenient experimentation.
23
 
24
- ## Contents
25
- The datasets are organized by common ML task type:
26
-
27
- ### Binary Classification
28
- Adult dataset
29
 
30
- Bank Marketing dataset
31
 
32
- ### Multiclassification
33
- Covertype dataset
34
 
35
- Statlog (Shuttle) dataset
36
-
37
- ### Regression
38
- Year Prediction MSD dataset
39
 
40
- ## Source and License
41
 
42
- All datasets in this repository are sourced from the UCI Machine Learning Repository and are used under **Creative Commons Attribution 4.0 International (CC BY 4.0)**.
 
 
 
43
 
44
- - CC BY 4.0 license text: https://creativecommons.org/licenses/by/4.0/
45
 
46
- ### Upstream dataset pages (UCI)
47
 
48
  - Adult (Census Income): https://archive.ics.uci.edu/dataset/2/adult
49
- - Bank Marketing: https://archive.ics.uci.edu/dataset/222/bank+marketing
50
- - Covertype: https://archive.ics.uci.edu/dataset/31/covertype
51
- - Statlog (Shuttle): https://archive.ics.uci.edu/dataset/148/statlog+shuttle
52
- - YearPredictionMSD: https://archive.ics.uci.edu/dataset/203/yearpredictionmsd
53
 
54
- ## Modifications and Data Processing Notes
55
- Some modifications may have been applied for usability and consistency. These modifications are intended to be non-substantive and not to change the meaning of the data.
56
 
57
- Typical changes include:
58
- - Adding **column headers** where the original files did not include headers.
59
- - Converting original formats (for example space-separated or other delimiters) into **`.csv`**.
60
- - Normalizing line endings and basic formatting fixes to improve parsing.
61
- - In some cases, reorganizing files into a standard folder structure (for example `train.csv` and `test.csv`).
62
 
63
- Unless explicitly stated in a dataset folder README (if present), no attempt was made to:
64
- - Remove rows or features
65
- - Alter feature values
66
- - Rebalance classes
67
- - Impute missing values
68
 
69
- If you require a byte-for-byte identical copy of the upstream distribution, please download directly from the corresponding UCI page.
70
 
71
- ## Missing Values
72
 
73
- Missing values are preserved as in the upstream sources. Depending on the dataset, missingness may appear as empty fields, `?`, or other dataset-specific markers. Refer to each dataset's UCI documentation for the authoritative description.
 
74
 
75
- ## Intended Use
76
 
77
- This pack is intended for:
78
- - Tabular model benchmarking (linear models, tree models, neural networks)
79
- - Privacy and security research on tabular learning pipelines, including:
80
- - reconstruction and gradient inversion attacks
81
- - defenses such as clipping, noise injection, discretization, constraint-aware decoding
82
- - evaluating reconstructibility using feature-level and record-level metrics
83
 
84
- It is not intended for making decisions about individuals or for any high-stakes deployment
 
1
  ---
2
+ license: other
3
  language:
4
  - en
5
+ pretty_name: LeakProTabular Public Benchmark Datasets
6
  task_categories:
7
  - tabular-classification
8
  - tabular-regression
9
  tags:
 
10
  - tabular
11
  - classification
12
  - binary
 
13
  - regression
14
+ - adult
15
+ - california-housing
16
+ - gradient-inversion
17
+ - federated-learning
18
+ - privacy
19
  size_categories:
20
  - 10K<n<100K
 
21
  ---
22
 
23
+ # LeakProTabular Public Benchmark Datasets
 
24
 
25
+ This repository provides public benchmark datasets used with the tabular-GIA research artifact for experiments on gradient inversion attacks in tabular federated learning.
 
 
 
 
26
 
27
+ For the experiment setup in tabular-GIA, Adult is downloaded from this Hugging Face repository, while California Housing is downloaded from the official scikit-learn source or here.
28
 
29
+ This repository does not include MIMIC-IV, the private multiclass benchmark, generated experiment outputs, checkpoints, plots, or attack reconstructions.
 
30
 
31
+ ## Contents
 
 
 
32
 
 
33
 
34
+ | Dataset | Task | Hosted path | Source |
35
+ |---|---|---|---|
36
+ | Adult | Binary classification | `binary/adult/` | UCI Machine Learning Repository |
37
+ | California Housing | Regression | `regression/california_housing/` | scikit-learn California Housing |
38
 
 
39
 
40
+ ## Dataset Sources
41
 
42
  - Adult (Census Income): https://archive.ics.uci.edu/dataset/2/adult
43
+ - California Housing: https://scikit-learn.org/stable/modules/generated/sklearn.datasets.fetch_california_housing.html
44
+ - Original California Housing reference: R. Kelley Pace and Ronald Barry. Sparse spatial autoregressions. Statistics & Probability Letters, 33(3):291-297, 1997.
 
 
45
 
46
+ ## Processing Notes
 
47
 
48
+ The hosted files are provided in CSV-compatible tabular form.
 
 
 
 
49
 
50
+ The tabular-GIA codebase applies its own preprocessing during experiments, including feature type handling, train/validation/test splitting, normalization, and model-specific categorical encoding.
 
 
 
 
51
 
52
+ ## Restricted and Private Data
53
 
54
+ This repository does not redistribute restricted or private datasets:
55
 
56
+ - MIMIC-IV requires credentialed PhysioNet access and must be obtained from PhysioNet.
57
+ - The private multiclass benchmark used in the paper cannot be shared.
58
 
59
+ ## Citation
60
 
61
+ If you use this repository with the accompanying code, please cite the associated paper after publication.
 
 
 
 
 
62
 
63
+ For the upstream datasets, cite the original dataset sources listed above.