| Change-point search | changepoint_search | Offline change-point detection on one series or several jointly: PELT, binary segmentation, bottom-up, window sliding, dynamic programming and kernel search over ten segment costs (mean, median, mean-and-covariance, kernel distribution, cosine, regression, continuous trend, rank, Mahalanobis, autoregressive), stopping by a number of breaks, a penalty or a cost budget; and the rolling change score of a sliding window. | 2 |
| Segmentation metrics | changepoint_metrics | Compare two sets of change points (Hausdorff distance, Rand index, Hamming distance, precision, recall and F1 within a margin, mean time error) and score given break dates against the data under a segment cost (each break’s gain against the BIC/AIC reference penalty). | 2 |
| Financial regimes and structural breaks | changepoint_regimes | Regime shifts in returns and volatility (one asset, or covariance and correlation regimes of several), trend changes of prices, rates and macro series, breaks in a regression (beta, hedge ratio) and in autoregressive dynamics (persistence, mean reversion of a spread), each with statistics per regime. | 4 |
| Penalty and model selection | changepoint_selection | How many breaks: the number of change points found at each penalty of a grid (with the widest stable plateau and the BIC/AIC reference penalty), and the total cost of the best segmentation with 0..K breaks scored by BIC, AIC or the elbow of the cost curve. | 2 |
| Synthetic teaching signals | changepoint_synthetic | Synthetic piecewise signals with known change points (constant means, Gaussian correlation switches, linear-model coefficients, sine frequencies) from a caller-chosen seed, for demonstrations and for checking how well a method and cost recover the true breaks. | 1 |
