stepp1/is_sparse_15d

---
license: mit
task_categories:
- tabular-classification
language:
- en
tags:
- synthetic
- sparse-learning
- classification
size_categories:
- 100K<n<1M
---

# is_sparse/sparse15d

## Dataset Description

This is a synthetic 15-dimensional classification dataset designed for sparse learning research.
The dataset contains 3 classes and is specifically designed to have sparse optimal representations,
where only a subset of features are informative for the classification task.

### Dataset Summary

- **Variant**: sparse15d
- **Features**: 15 continuous features
- **Classes**: 3
- **Entropy(Y)**: 1.4855
- **Mutual Information (joint)**: 1.1819
- **Maximum Achievable Accuracy**: 0.8967

## Dataset Structure

### Data Instances

Each instance consists of:
- `data`: A 15-dimensional feature vector (float32)
- `label`: An integer class label (0, 1, or 2)

### Data Splits

| Split | Number of Instances |
|-------|---------------------|
| Train | Variable (see below) |
| Test | Variable (see below) |

## Dataset Creation

This dataset was synthetically generated for research on sparse learning and optimal feature selection.
The mutual information values between feature subsets and labels are provided in the metadata.

### Mutual Information Structure

The dataset includes ground-truth mutual information values for various feature subsets, enabling:
- Feature importance analysis
- Information-theoretic learning algorithms
- Benchmarking of MI estimation methods

Key MI values:
- joint: 1.1819
- 1: 0.3273
- 1-2: 0.3273
- 1-2-3: 0.6634
- 1-2-3-4: 0.6634
- 1-2-3-4-5: 1.1819
- 1-2-3-5: 1.1819
- 1-2-4: 0.3273
- 1-2-4-5: 1.0492
- 1-2-5: 1.0492

## Citation

If you use this dataset, please cite the associated research paper (to be added).

## License

MIT License