Rust implementation of ROCKET time-series transform with Python bindings

This repo implements the ROCKET time-series transform in Rust with Python bindings. Originally, ROCKET is implemented in Python.

Usage

Using the arrow-head dataset provided by aeon:

from aeon.datasets import load_arrow_head
from rocket_rs import transform

x, _ = load_arrow_head(split="train")
z = transform(x)

Contributing

Initial setup

• Install Python, e.g. using miniforge
• Install Rust
• Install maturin
• Create and activate Python virtual environment (e.g. conda create -n rocket-rs python=3.11 && conda activate rocket-rs)
• Run:

maturin develop --extras=dev
pre-commit install --install-hooks
python -m maturin_import_hook site install --args="--release"

Development

• maturin develop to re-build and install the editable development version (pip install --editable .)
• maturin build to build the Rust package and Python wheel
• pytest --benchmark-skip to rebuild automatically using maturin_import_hook and run Python unit tests
• pytest --benchmark-only to run benchmarks using maturin_import_hook

Todo

• separate kernel parameter generation from computation
• compare and test against Python implementation in aeon
• refactor to pass random generator to functions
• use debugger
• benchmark & improve

Notes

• For Python bindings for Rust, see https://github.com/PyO3/pyo3
• https://github.com/PyO3/maturin for packaging and publishing
• To use poetry with maturin, see PyO3/maturin#1246 (comment)
• For project layout using both Python and Rust, see https://www.maturin.rs/project_layout.html#mixed-rustpython-project, I preferred having separate folders for Rust and Python
• To generate GitHub Action, run maturin generate-ci github
• https://github.com/FL33TW00D/rustDTW
• https://github.com/PyO3/rust-numpy for integration between Python Numpy and Rust ndarray arrays
• https://www.aeon-toolkit.org/en/stable/examples/classification/classification.html
• https://github.com/projectsyn/reclass-rs and https://www.youtube.com/watch?v=N7GMHcX-WdA
• micro-benchmarking with https://github.com/bheisler/criterion.rs