Crate tsai_models 

tsai_models

Model zoo for tsai-rs: CNN, Transformer, ROCKET, RNN, and Tabular architectures.

This crate provides deep learning architectures for time series:

CNN Models

• InceptionTimePlus - InceptionTime with improvements
• ResNetPlus - ResNet adapted for time series
• [XceptionTimePlus] - Xception-inspired architecture
• OmniScaleCNN - Multi-scale CNN
• XCMPlus - Explainable CNN

Transformer Models

• TSTPlus - Time Series Transformer
• TSiTPlus - Improved Time Series Transformer with multiple PE options
• TSPerceiver - Perceiver for time series
• PatchTST - Patch-based Transformer

ROCKET Family

• MiniRocket - Fast random convolutional features
• [MultiRocketPlus] - Multiple ROCKET kernels
• HydraPlus - Hybrid ROCKET

RNN Models

• RNNPlus - LSTM/GRU with improvements
• RNNAttention - RNN with attention

Hybrid Models

• RNNFCN - RNN-FCN hybrid (LSTM-FCN, GRU-FCN)

Tabular Models

• TabTransformer - Transformer for tabular data
• TabFusionTransformer - Fusion of time series and tabular

Re-exports

pub use checkpoint::save_model;

pub use checkpoint::load_record;

pub use checkpoint::CheckpointError;

pub use checkpoint::CheckpointFormat;

pub use checkpoint::CheckpointMetadata;

pub use checkpoint::CheckpointPrecision;

pub use checkpoint::ModelCheckpoint;

pub use registry::default_registry;

pub use registry::ModelRegistry;

pub use registry::RegistryError;

pub use registry::TSModel;

pub use cnn::*;

pub use hybrid::*;

pub use rnn::*;

pub use rocket::*;

pub use tabular::*;

pub use transformer::*;

Modules

checkpoint

Model checkpointing and serialization utilities.

cnn

CNN models for time series.

hybrid

Hybrid models combining different architectures.

registry

Model registry for dynamic model creation.

rnn

RNN models for time series.

rocket

ROCKET family models.

tabular

Tabular models for time series with structured data.

traits

Trait implementations for model training.

transformer

Transformer models for time series.