The 1000×1000 collection: 1000 synthetic time series from 133 dynamical processes

The 1000×1000 collection contains 1000 simulated time series, each 1000 samples long, generated from 133 mechanisms across 13 classes: noise and linear processes, long-memory and multifractal processes, nonlinear stochastic processes, deterministic maps, chaotic flows, oscillators, noise-driven continuous-time dynamics, point processes and discrete states, composite and coupled systems, observation effects, and models of hearts, brains, climate and ecosystems.

All series are stationary, real-valued and finite-variance.</p><p dir="ltr">Each series comes with its generating process, drawn parameters, and property tags (e.g., chaotic, long memory, heavy-tailed, bursty). The collection is a rethinking of the existing Empirical1000 dataset, in the same <i>hctsa</i> file formats, and includes the full <i>hctsa</i> feature matrix (v3.0, 7077 features).</p><p dir="ltr">Files: INP_1000x1000.mat (<i>hctsa</i> input), HCTSA_1000x1000.mat (hctsa features), hctsa_*.csv (the same as CSV), metadata.csv (ground truth and tags), series.npz (NumPy).