- About table: complexity of the corrected sampler (block-mean reduction beyond 2000
samples) and its reference implementation (pyhsmm); "zhd96" is an unrelated HDP-HMM.
- Drop-down choices for the new string options dur_family, block_features and init.
fix(app): neutral coloring when no CP found + editable n_segments for CLaSP/Clap (0 = auto)
- Add Home tab with logo, project overview, demo screenshot
- Rename 'MoCap (built-in)' -> 'Sample dataset', display first by default
- Synthetic generator: new Gaussian-process pattern (default), continuous=True
to suppress segment-boundary discontinuities
- 'Use GT' button now also fills n_states, n_clusters, n_regimes, k_max,
n_segments; GT-like params reordered to top of the panel
- Each algorithm header shows per-method asymptotic complexity in math
notation, colorised by tier (linear / log-linear / quadratic / cubic),
with tooltip + 'Complexity notation' accordion explaining n, d, l, n_cp, n_s
- Algorithm families: drop 'Other' fallback, lookup via curated ALGO_BY_NAME
(fixes InformationGain etc.)
- About tab: replace Status column with Complexity, drop group separator rows
- Detectors: enable Beast via vendored Rbeast; exclude SNLDS / ChangeFinder;
GreedyGaussian correctly tagged change_point_detection only
- requirements.txt: document optional Rbeast dependency
- ru…
Only CLaP and AutoPlait were available. Now install selected tsseg
extras to enable deep learning, prophet, aeon-based and numba-accelerated
algorithms on HuggingFace Spaces.
