pmtvs 1.0.2

Rust-accelerated signal analysis primitives

pmtvs

Rust-accelerated signal analysis primitives.

numpy in, number out.

Installation

pip install pmtvs

Quick Start

from pmtvs import hurst_exponent
import numpy as np

signal = np.random.randn(1000)
h = hurst_exponent(signal)
print(f"Hurst exponent: {h:.3f}")

import numpy as np
from pmtvs import (
    sample_entropy, permutation_entropy,
    hurst_exponent, dfa,
    skewness, kurtosis,
    eigendecomposition, svd_decomposition,
    granger_causality, mutual_information,
)

signal = np.random.randn(5000)

# Entropy
print(sample_entropy(signal))       # ~2.19
print(permutation_entropy(signal))  # ~2.57

# Fractal
print(hurst_exponent(signal))       # ~0.50
print(dfa(signal))                  # ~0.50

# Statistics
print(skewness(signal))             # ~0.00
print(kurtosis(signal))             # ~3.00

What's Inside

232 functions in one flat import. Single wheel, no sub-packages.

Domain	Functions	Description
Entropy	6	Sample entropy, permutation entropy, LZ complexity
Fractal	7	Hurst exponent, DFA, rescaled range, variance growth
Statistics	34	Descriptive stats, calculus, derivatives, normalization
Correlation	15	Autocorrelation, cross-correlation, Spearman, Kendall, coherence
Distance	6	Euclidean, cosine, Manhattan, DTW, EMD
Embedding	5	Time delay embedding, FNN, Cao's method
Coupling	2	Rolling Spearman correlation
Dynamics	23	Lyapunov exponents, RQA, attractor reconstruction
Spectral	17	FFT (rustfft), PSD, Hilbert transform
Matrix	22	SVD, covariance, DMD, MI/Granger matrices
Topology	11	Persistent homology, Betti numbers
Network	10	Graph centrality, community detection
Information	15	Mutual information, KL divergence, Granger causality
Physics	1	Subspace angle
Tests	stat tests	Stationarity, hypothesis testing
Regression	5	Linear regression, signal arithmetic

Full mathematical reference with LaTeX equations: PRIMITIVES.md

Benchmarks

Measured on Mac Mini M4 (arm64), Python 3.12. Not theoretical — timed.

Function	Speedup	Why
dfa	267x	Rust eliminates per-segment polyfit overhead
dtw_distance	94x	O(n^2) — Rust loops crush Python loops
sample_entropy	31x	O(n^2) template matching
permutation_entropy	28x	Combinatorial pattern counting
hurst_exponent	20x	Hybrid Rust/numpy by signal size
skewness	7x	Single-pass Rust vs multi-pass numpy
kurtosis	8x	Same
autocorrelation	4x
euclidean_distance	1.6x	numpy BLAS — at parity
rms	~1x	numpy BLAS — at parity
cosine_distance	~1x	numpy BLAS — at parity

Overall: 41x mean, 326x max.

BLAS-backed ops (rms, cosine, euclidean) delegate to numpy when numpy's LAPACK/BLAS is faster. Algorithmic functions (loops, O(n^2), combinatorial) use Rust. No ego — just speed.

Run the benchmark yourself:

python benchmarks/rust_vs_python.py

Development

git clone https://github.com/pmtvs/pmtvs.git
cd pmtvs

# Install (requires Rust toolchain + maturin)
pip install maturin numpy scipy
cd packages/pmtvs && maturin develop --release

# Run tests
pytest packages/*/tests/ -v

Reporting Issues

The most useful bug report includes: a minimal code example, the value pmtvs returned, and the value you expected with a source (published paper, competing library, or analytical solution).

"sample_entropy returns 2.14 but Richman-Moorman Table 2 gives 2.09 for this test vector" is a perfect bug report.

Edge cases are especially welcome — constant signals, very short signals, NaN-heavy data, extreme outliers. If you can break it, we want to know.

Issues: [e-mail]

Citation

@software{pmtvs2026,
  author    = {Rudder, Jason},
  title     = {pmtvs: Rust-Accelerated Signal Analysis Primitives},
  year      = {2026},
  publisher = {PyPI},
  url       = {https://pypi.org/project/pmtvs/}
}

Rudder Research © 2026

License

PolyForm Strict 1.0.0 with Additional Terms.

• Students & individual researchers: Free. Cite us.
• Funded research labs (grants > $100K): Academic Research License required.
• Commercial use: Commercial License required.
• Institutional deployment: Institutional License required.

Contact: [e-mail]

See LICENSE for full terms.