Skip to content

API Reference

Welcome to the Spectrans API reference documentation. This section provides detailed information about all modules, classes, and functions in the Spectrans library.

Core Modules

Transforms

Spectral transform implementations including FFT, wavelet, cosine, and Hadamard transforms.

Kernels

Kernel approximation methods for efficient attention computation.

Layers

Neural network layers including attention mechanisms, mixing layers, and operators.

Blocks

Pre-built transformer blocks combining attention and feedforward components.

Models

Complete model architectures for various tasks.

Utils

Utility functions for complex operations, padding, and initialization.

Core

Core interfaces, base classes, and registry system.

Configuration

Configuration system for models and components.

Module Organization

The library is organized into the following main packages:

spectrans/
├── core/           # Core interfaces and registry
│   ├── base        # Base classes
│   ├── types       # Type definitions
│   └── registry    # Component registry
├── transforms/     # Spectral transforms
│   ├── base        # Base transform classes
│   ├── fourier     # FFT implementations
│   ├── cosine      # DCT/DST transforms
│   ├── hadamard    # Hadamard transforms
│   └── wavelet     # Wavelet transforms
├── kernels/        # Kernel approximations
│   ├── base        # Base kernel classes
│   ├── rff         # Random Fourier Features
│   └── spectral    # Spectral kernels
├── layers/         # Neural network layers
│   ├── attention/  # Attention mechanisms
│   │   ├── lst     # Linear Spectral Transform
│   │   └── spectral # Spectral attention
│   ├── mixing/     # Mixing layers
│   │   ├── base    # Base mixing classes
│   │   ├── afno    # AFNO mixing
│   │   ├── fourier # Fourier mixing
│   │   ├── global_filter # Global filters
│   │   └── wavelet # Wavelet mixing
│   └── operators/  # Neural operators
│       └── fno     # Fourier Neural Operators
├── blocks/         # Transformer blocks
│   ├── base        # Base block classes
│   ├── hybrid      # Hybrid blocks
│   └── spectral    # Spectral blocks
├── models/         # Complete architectures
│   ├── base        # Base model classes
│   ├── afno        # AFNO models
│   ├── fnet        # FNet models
│   ├── gfnet       # GFNet models
│   └── ...         # Other models
├── utils/          # Utility functions
│   ├── complex     # Complex operations
│   ├── initialization # Initialization
│   └── padding     # Padding utilities
└── config/         # Configuration system
    ├── core        # Core configuration
    ├── builder     # Model builder
    ├── models      # Model configs
    └── layers/     # Layer configs
        ├── attention # Attention configs
        └── mixing   # Mixing configs

Mathematical Notation

Throughout the documentation, we use standard mathematical notation:

  • Vectors are denoted by lowercase bold letters: \(\mathbf{x}\), \(\mathbf{y}\)
  • Matrices are denoted by uppercase bold letters: \(\mathbf{W}\), \(\mathbf{K}\)
  • Scalars are denoted by lowercase letters: \(n\), \(d\), \(k\)
  • Functions and operators use calligraphic letters: \(\mathcal{G}\), \(\mathcal{F}\)
  • Complex numbers use \(\mathbb{C}\), real numbers use \(\mathbb{R}\)

API Stability

The Spectrans API follows semantic versioning. Classes and functions marked as stable will maintain backward compatibility within major versions. Experimental features are clearly marked in the documentation.