tsfast 0.5.0

TSFast

Description

tsfast is an open-source deep learning library for time series analysis and system identification tasks. Built on PyTorch, it offers efficient deep learning models and utilities.

tsfast is an open-source deep learning package that focuses on system identification and time series analysis tasks. Built on PyTorch, it provides efficient implementations of various deep learning models and utilities.

Key Features

• Specialized Data Handling — HDF5-backed data pipeline with sequence transforms, noise injection, normalization, and TBPTT (Truncated Backpropagation Through Time) support.
• Benchmark Datasets — One-line access to Silverbox, Wiener-Hammerstein, Cascaded Tanks, and more via identibench.
• Time Series Models — RNNs (DenseNet, Residual), TCNs, CRNNs, transformers, NARX, state-space models (NeuralStateSpace, S5, LRU, Mamba), dynoNet, port-Hamiltonian networks (PHNN), and certified-by-construction REN/R2DN models, with fused Triton/C++ kernels for the recurrent architectures.
• Integrated Training — RNNLearner, TCNLearner, CRNNLearner with custom losses (nrmse, cut_loss) and composable transforms.
• System Identification — Simulation, N-step prediction, FranSys, and autoregressive models with InferenceWrapper for deployment.
• Physics-Informed NNs — Embed governing equations into training via CollocationLoss, PhysicsLoss, and PIRNN, or train physics-only surrogates without measured data (DD-PINN, PINC).
• Hyperparameter Optimization — Ray Tune integration via HPOptimizer.
• Deployment — Model save/load, InferenceWrapper for NumPy-in/NumPy-out inference, and ONNX export.
• Performance — CUDA graph acceleration via GraphedStatefulModel for low-overhead GPU training.

Installation

You can install the latest stable version using:

pip install tsfast

For development installation:

git clone https://github.com/daniel-om-weber/tsfast
cd tsfast
pip install -e '.[dev]'
# or using uv:
uv sync --extra dev

Quick Start

Here is a quick example using a benchmark dataloader. It demonstrates loading and visualizing data, training a RNN, and visualizing the results.

from tsfast.basics import *

# Load benchmark dataset
dls = create_dls_silverbox()

# Train an RNN and visualize results
lrn = RNNLearner(dls)
lrn.fit_flat_cos(1)
lrn.show_results(max_n=1)

Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

Citation

If you use tsfast in your research, please cite:

@Misc{tsfast,
author = {Daniel O.M. Weber},
title = {tsfast - A deep learning library for time series analysis and system identification},
howpublished = {Github},
year = {2024},
url = {https://github.com/daniel-om-weber/tsfast}
}