'MiniRocket': A Very Fast (Almost) Deterministic Transform for Time Series Classification

High-performance R and C++ implementation using 'OpenMP' parallelization for the 'MiniRocket' algorithm. Extracts features from univariate time series for downstream classification as described in 'Dempster et al.' (2021) ().

minirocketR

minirocketR provides a high-performance R implementation of the MiniRocket time series feature extraction algorithm (Dempster et al., 2021). Built on a native C++ core via Rcpp with OpenMP multithreading support, the package delivers exact mathematical parity with the Python reference implementation (sktime).

Key Features

• Mathematical Parity: Implements alternating same/valid zero-padded convolutions, golden-ratio quasi-random quantile sampling, and exact dilation weighting.
• C++ Performance: Optimized core routines for fast execution on large time series datasets.
• OpenMP Parallelization: Multi-core acceleration configurable via the num_threads argument.
• PPV Feature Mapping: Standardized transformation yielding $9,996$ Proportion of Positive Values (PPV) features bounded in $[0, 1]$.

Installation

# Install from local source
install.packages("minirocketR")