Module advanced_training 

Advanced Training Methods for Time Series

This module has been refactored into a modular structure for better maintainability and organization. All original functionality is preserved through re-exports.

Refactored Module Structure

The advanced training functionality has been organized into focused sub-modules:

• Configuration: Common data structures and configurations
• Meta-Learning: MAML and other meta-learning algorithms
• Neural ODEs: Continuous-time neural networks with ODE solvers
• Variational Methods: VAEs for probabilistic time series modeling
• Transformers: Attention-based sequence modeling
• Hyperparameter Optimization: Automated parameter tuning
• Few-Shot Learning: Prototypical Networks and REPTILE
• Memory-Augmented Networks: External memory architectures
• Meta-Optimization: Learned optimizers for adaptive updates

Usage

All original APIs are preserved. You can continue to use this module exactly as before:

use scirs2_series::advanced_training::{MAML, TimeSeriesVAE, TimeSeriesTransformer};

// Create a MAML instance
let mut maml = MAML::<f64>::new(4, 8, 2, 0.01, 0.1, 5);

// Create a VAE for time series
let vae = TimeSeriesVAE::<f64>::new(10, 3, 5, 16, 16);

// Create a transformer for forecasting
let transformer = TimeSeriesTransformer::<f64>::new(12, 6, 64, 8, 4, 256);

Advanced Usage

For more specific functionality, you can also import directly from sub-modules:

use scirs2_series::advanced_training::few_shot::{PrototypicalNetworks, REPTILE};
use scirs2_series::advanced_training::hyperparameter_optimization::{
    HyperparameterOptimizer, OptimizationMethod
};

Modules

config

Configuration types for advanced training methods

few_shot

Few-Shot Learning Algorithms

hyperparameter_optimization

Hyperparameter Optimization Framework

memory_augmented

Memory-Augmented Neural Networks

meta_learning

Meta-learning algorithms for few-shot time series forecasting

neural_ode

Neural Ordinary Differential Equations for continuous-time modeling

optimization

Meta-Optimization Algorithms

transformers

Transformer-based Time Series Forecasting

variational

Variational Autoencoder for Time Series

Structs

FewShotEpisode

Few-shot learning episode data structure

HyperparameterOptimizer

Hyperparameter Optimization Framework

HyperparameterSet

Set of hyperparameters

MAML

Model-Agnostic Meta-Learning (MAML) for few-shot time series forecasting

MANN

Memory-Augmented Neural Network (MANN)

MetaOptimizer

Meta-Optimizer using LSTM to generate parameter updates

NeuralODE

Neural Ordinary Differential Equation (NODE) implementation

ODESolverConfig

Configuration for ODE solver

OptimizationProblem

Optimization problem for meta-optimizer training

OptimizationResults

Optimization results

OptimizationStep

Single optimization step

PrototypicalNetworks

Prototypical Networks for Few-Shot Learning

REPTILE

REPTILE Algorithm for Meta-Learning

SearchSpace

Search space for hyperparameters

TaskData

Task data structure for meta-learning

TimeSeriesTransformer

Transformer model for time series forecasting with multi-head attention

TimeSeriesVAE

Variational Autoencoder for Time Series with Uncertainty Quantification

VAEOutput

VAE output structure

Enums

IntegrationMethod

Integration methods for ODE solving

OptimizationMethod

Hyperparameter optimization methods