SciRS2 Transform

Production-ready data transformation library for machine learning in Rust

This crate provides comprehensive data transformation utilities for the SciRS2 ecosystem, designed to match and exceed the functionality of scikit-learn's preprocessing module while leveraging Rust's performance and safety guarantees.

🚀 Features

Data Normalization & Standardization

• Min-Max Scaling: Scale features to [0, 1] or custom ranges
• Z-Score Standardization: Transform to zero mean and unit variance
• Robust Scaling: Use median and IQR for outlier-resistant scaling
• Max Absolute Scaling: Scale by maximum absolute value
• L1/L2 Normalization: Vector normalization

Feature Engineering

• Polynomial Features: Generate polynomial and interaction features
• Power Transformations: Box-Cox and Yeo-Johnson with optimal λ estimation
• Discretization: Equal-width and equal-frequency binning
• Binarization: Convert continuous features to binary
• Log Transformations: Logarithmic feature scaling

Dimensionality Reduction

• PCA: Principal Component Analysis with centering/scaling options
• Truncated SVD: Memory-efficient singular value decomposition
• LDA: Linear Discriminant Analysis for supervised reduction
• t-SNE: Advanced non-linear embedding with Barnes-Hut optimization

Categorical Encoding

• One-Hot Encoding: Sparse and dense representations
• Ordinal Encoding: Label encoding for ordinal categories
• Target Encoding: Supervised encoding with regularization
• Binary Encoding: Memory-efficient encoding for high-cardinality features

Missing Value Imputation

• Simple Imputation: Mean, median, mode, and constant strategies
• KNN Imputation: K-nearest neighbors with multiple distance metrics
• Iterative Imputation: MICE algorithm for multivariate imputation
• Missing Indicators: Track which values were imputed

Feature Selection

• Variance Threshold: Remove low-variance features
• Integration: Seamless integration with all transformers

📦 Installation

Add this to your Cargo.toml:

[dependencies]
scirs2-transform = "0.1.0-alpha.6"

For parallel processing and enhanced performance:

[dependencies]
scirs2-transform = { version = "0.1.0-alpha.6", features = ["parallel"] }

🎯 Quick Start

Basic Normalization

use ndarray::array;
use scirs2_transform::normalize::{normalize_array, NormalizationMethod, Normalizer};

// One-shot normalization
let data = array![[1.0, 2.0, 3.0], 
                 [4.0, 5.0, 6.0],
                 [7.0, 8.0, 9.0]];
let normalized = normalize_array(&data, NormalizationMethod::MinMax, 0)?;

// Fit-transform workflow for reusable transformations
let mut normalizer = Normalizer::new(NormalizationMethod::ZScore, 0);
let train_transformed = normalizer.fit_transform(&train_data)?;
let test_transformed = normalizer.transform(&test_data)?;

Feature Engineering

use scirs2_transform::features::{PolynomialFeatures, PowerTransformer, binarize};

// Generate polynomial features
let data = array![[1.0, 2.0], [3.0, 4.0]];
let poly = PolynomialFeatures::new(2, false, true);
let poly_features = poly.transform(&data)?;

// Power transformations with optimal lambda
let mut transformer = PowerTransformer::yeo_johnson(true);
let gaussian_data = transformer.fit_transform(&skewed_data)?;

// Binarization
let binary_features = binarize(&data, 0.0)?;

Dimensionality Reduction

use scirs2_transform::reduction::{PCA, TSNE};

// PCA for linear dimensionality reduction
let mut pca = PCA::new(2, true, false);
let reduced_data = pca.fit_transform(&high_dim_data)?;
let explained_variance = pca.explained_variance_ratio().unwrap();

// t-SNE for non-linear visualization
let mut tsne = TSNE::new(2, 30.0, 500)?;
let embedding = tsne.fit_transform(&data)?;

Categorical Encoding

use scirs2_transform::encoding::{OneHotEncoder, TargetEncoder};

// One-hot encoding
let mut encoder = OneHotEncoder::new(false, false)?;
let encoded = encoder.fit_transform(&categorical_data)?;

// Target encoding for supervised learning
let mut target_encoder = TargetEncoder::mean_encoding(1.0);
let encoded = target_encoder.fit_transform(&categories, &targets)?;

Missing Value Imputation

use scirs2_transform::impute::{SimpleImputer, KNNImputer, ImputeStrategy};

// Simple imputation
let mut imputer = SimpleImputer::new(ImputeStrategy::Mean);
let complete_data = imputer.fit_transform(&data_with_missing)?;

// KNN imputation
let mut knn_imputer = KNNImputer::new(5)?;
let imputed_data = knn_imputer.fit_transform(&data_with_missing)?;

🔧 Advanced Usage

Pipeline Integration

// Sequential transformations
let mut scaler = Normalizer::new(NormalizationMethod::ZScore, 0);
let mut pca = PCA::new(50, true, false);

// Preprocessing pipeline
let scaled_data = scaler.fit_transform(&raw_data)?;
let reduced_data = pca.fit_transform(&scaled_data)?;

Custom Transformations

use scirs2_transform::features::PowerTransformer;

// Custom power transformation
let mut transformer = PowerTransformer::new("yeo-johnson", true)?;
transformer.fit(&training_data)?;

// Apply to new data
let transformed_test = transformer.transform(&test_data)?;
let original_test = transformer.inverse_transform(&transformed_test)?;

Performance Optimization

// Enable parallel processing for large datasets
use rayon::prelude::*;

// Most transformers automatically use parallel processing when beneficial
let mut pca = PCA::new(100, true, false);
let result = pca.fit_transform(&large_dataset)?; // Automatically parallelized

📊 Performance

SciRS2 Transform is designed for production workloads:

• Memory Efficient: Zero-copy operations where possible
• Parallel Processing: Multi-core support via Rayon
• SIMD Ready: Integration with vectorized operations
• Large Scale: Handles datasets with 100k+ samples and 10k+ features

Benchmarks

🧪 Testing

Run the comprehensive test suite:

# All tests (100 tests)
cargo test

# With output
cargo test -- --nocapture

# Specific module
cargo test normalize::tests

📚 Documentation

• API Documentation
• SciRS2 Project
• Performance Guide
• Migration from sklearn

🔄 Compatibility

Scikit-learn API Compatibility

SciRS2 Transform follows scikit-learn's API conventions:

• fit() / transform() / fit_transform() pattern
• Consistent parameter naming
• Similar default behaviors
• Compatible data formats (via ndarray)

Migration from Scikit-learn

# Python (scikit-learn)
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

// Rust (SciRS2)
use scirs2_transform::normalize::{Normalizer, NormalizationMethod};
let mut scaler = Normalizer::new(NormalizationMethod::ZScore, 0);
let x_scaled = scaler.fit_transform(&x)?;

🏗️ Architecture

Modular Design

scirs2-transform/
├── normalize/     # Data normalization and standardization
├── features/      # Feature engineering utilities  
├── reduction/     # Dimensionality reduction algorithms
├── encoding/      # Categorical data encoding
├── impute/        # Missing value imputation
├── selection/     # Feature selection methods
└── scaling/       # Advanced scaling transformers

Error Handling

Comprehensive error handling with descriptive messages:

use scirs2_transform::{Result, TransformError};

match normalizer.fit_transform(&data) {
    Ok(transformed) => println!("Success!"),
    Err(TransformError::InvalidInput(msg)) => println!("Input error: {}", msg),
    Err(TransformError::TransformationError(msg)) => println!("Transform error: {}", msg),
    Err(e) => println!("Other error: {}", e),
}

🛠️ Development

Building from Source

git clone https://github.com/cool-japan/scirs
cd scirs/scirs2-transform
cargo build --release

Contributing

1. Fork the repository
2. Create a feature branch
3. Add comprehensive tests
4. Ensure all tests pass: cargo test
5. Run clippy: cargo clippy
6. Submit a pull request

📈 Roadmap

Version 0.1.0 (Next)

• Pipeline API for chaining transformations
• Enhanced SIMD acceleration
• Comprehensive benchmarking suite

Version 0.2.0

• UMAP and manifold learning algorithms
• Advanced matrix decomposition methods
• Time series feature extraction

Version 1.0.0

• GPU acceleration
• Distributed processing
• Production monitoring tools

🤝 License

This project is dual-licensed under either:

• MIT License
• Apache License Version 2.0

You may choose to use either license.

🙏 Acknowledgments

• Inspired by scikit-learn
• Built on ndarray
• Powered by Rayon for parallelization

Ready for Production: SciRS2 Transform v0.1.0-alpha.6 provides production-ready data transformation capabilities with performance that meets or exceeds established Python libraries while offering Rust's safety and performance guarantees.