datasets

Dataset Loading Utilities

Dataset loading utilities.

class SToG.datasets.DatasetLoader

Bases: object

Load and prepare datasets for benchmarking.

static load_breast_cancer()

Load breast cancer dataset.

Returns:

Dictionary with dataset information

static load_wine()

Load wine dataset.

Returns:

Dictionary with dataset information

static create_synthetic_high_dim()

Create synthetic high-dimensional dataset (MADELON-like).

Returns:

Dictionary with dataset information

static create_synthetic_correlated()

Create synthetic dataset with correlated features.

Returns:

Dictionary with dataset information

DatasetLoader

class SToG.datasets.DatasetLoader

Bases: object

Load and prepare datasets for benchmarking.

static load_breast_cancer()

Load breast cancer dataset.

Returns:

Dictionary with dataset information

static load_wine()

Load wine dataset.

Returns:

Dictionary with dataset information

static create_synthetic_high_dim()

Create synthetic high-dimensional dataset (MADELON-like).

Returns:

Dictionary with dataset information

static create_synthetic_correlated()

Create synthetic dataset with correlated features.

Returns:

Dictionary with dataset information

Overview

The SToG.datasets.DatasetLoader provides static methods to load benchmark datasets suitable for feature selection experiments.

Available Datasets

All loaders return a dictionary with:

{
    'name': str,              # Dataset name
    'X': ndarray,             # Feature matrix [n_samples, n_features]
    'y': ndarray,             # Target labels [n_samples]
    'n_important': int,       # Number of truly informative features
    'description': str        # Dataset description
}

Breast Cancer Dataset

static DatasetLoader.load_breast_cancer()

Load breast cancer dataset.

Returns:

Dictionary with dataset information

Properties:

• 569 samples

• 30 features

• Binary classification (malignant vs. benign)

• ~10 informative features

• Real-world dataset (UCI repository)

Use case: General feature selection testing

from SToG import DatasetLoader
data = DatasetLoader.load_breast_cancer()
print(f"Dataset: {data['name']}")
print(f"Shape: {data['X'].shape}")
print(f"Informative features: {data['n_important']}")

Wine Dataset

static DatasetLoader.load_wine()

Load wine dataset.

Returns:

Dictionary with dataset information

Properties:

• 178 samples

• 13 features

• 3-class classification

• ~7 informative features

• Real-world dataset (UCI repository)

Use case: Multi-class feature selection

data = DatasetLoader.load_wine()
print(f"Classes: {len(np.unique(data['y']))}")

Synthetic High-Dimensional Dataset

static DatasetLoader.create_synthetic_high_dim()

Create synthetic high-dimensional dataset (MADELON-like).

Returns:

Dictionary with dataset information

Properties:

• 600 samples

• 100 features

• Binary classification

• 5 informative features

• 10 redundant features

• 85 noise features

• Class imbalance: ~3%

Use case: High-dimensional feature selection testing

Generated via: scikit-learn’s make_classification

data = DatasetLoader.create_synthetic_high_dim()
print(f"Sparsity level: {5}/{data['X'].shape[1]}")

Synthetic Correlated Features Dataset

static DatasetLoader.create_synthetic_correlated()

Create synthetic dataset with correlated features.

Returns:

Dictionary with dataset information

Properties:

• 500 samples

• 50 features

• Binary classification

• 5 core informative features

• 10 correlated copies (2 per core feature)

• 35 noise features

• Target: \(y = \mathbb{1}[x_1 + x_2 \cdot x_3 > 0]\)

Use case: Testing with correlated features

Characteristics:

• Features 0-4: core informative

• Features 5-14: noisy copies of core features

• Features 15-49: pure noise

• High correlation: 0.9+ between core and copies

data = DatasetLoader.create_synthetic_correlated()
# Use with CorrelatedSTGLayer for best results

Full Benchmarking Example

from SToG import DatasetLoader, ComprehensiveBenchmark

loader = DatasetLoader()
datasets = [
    loader.load_breast_cancer(),
    loader.load_wine(),
    loader.create_synthetic_high_dim(),
    loader.create_synthetic_correlated(),
]

benchmark = ComprehensiveBenchmark()
benchmark.run_benchmark(datasets)

Data Preprocessing

Recommended preprocessing pipeline:

from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split

data = DatasetLoader.load_breast_cancer()
X = data['X']
y = data['y']

# 1. Standardize features
scaler = StandardScaler()
X = scaler.fit_transform(X)

# 2. Split data
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)
X_train, X_val, y_train, y_val = train_test_split(
    X_train, y_train, test_size=0.2, random_state=42
)

# 3. Convert to tensors
X_train = torch.FloatTensor(X_train)
y_train = torch.LongTensor(y_train)