pyssaBSS 0.1.4

Stationary Subspace Analysis for temporal and spatial data

pyssaBSS

Stationary Subspace Analysis (SSA) for temporal and spatial data.

pyssaBSS implements methods for separating stationary and nonstationary subspaces in multivariate data, based on the joint eigen-decomposition of scatter operators. It provides SSA for time series and SPSSA for spatial fields.

Note that the C++ code used for joint diagonalization is taken from the CRAN R-package JADE under the GPL license. The original package can be found at: https://cran.r-project.org/package=JADE

---

Installation

pip install pyssaBSS

Pre-built wheels are available for Windows, macOS, and Linux (Python 3.8–3.13). No compiler required.

Building from source

git clone https://github.com/perttusaarela/pyssaBSS
cd pyssaBSS
pip install -e ".[dev]"

Requires CMake ≥ 3.15 and a C++14 compiler.

---

Quick example

import numpy as np
from pyssaBSS import SPSSA_COMB
from pyssaBSS.kernels import ScaledBallKernel
from pyssaBSS.spatial import partition_coordinates
# data: (n_signals, n_samples), coords: (n_samples, 2)
data   = np.random.randn(5, 400)
coords = np.random.uniform(0, 20, (400, 2))

# Partition the spatial domain
partition = partition_coordinates(coords, 3, 3, 20)

# Add nonstationarity in the mean to the last signal
for part in partition:
    data[-1, part] += np.random.uniform(low=-5, high=5)

# Decompose
model = SPSSA_COMB(data, coords, partition=partition, kernel=ScaledBallKernel(2.2))

# Estimate rank of non-stationary subspace
q = model.estimate_rank()
print(f"Estimated rank: {q}")

# Extract subspaces
stationary, non_stationary = model.subspaces(q)
print(f"Stationary subspace:     {stationary.shape}")
print(f"Non-stationary subspace: {non_stationary.shape}")

---

Overview

Class	Data type	Varies across
SSA	Time series	Time partitions
SPSSA	Spatial fields	Spatial segments

Convenience constructors (SSA_SIR, SPSSA_SIR, SPSSA_LCOR, SPSSA_COMB, ...) bundle a scatter operator and rank estimator into a single call.

---

License

GLP 3.0 — see LICENSE.txt.