hdMTD 0.1.5

FSC method

Corrected the sample split in hdMTD_FSC() so that FS is applied to the chronologically older observations and CUT to the most recent observations, in accordance with the theoretical procedure.
Added the cut_fraction argument to control the proportion of the sample allocated to CUT. Its default value is 0.5.
Added validation ensuring that cut_fraction is strictly between 0 and 1 and that both resulting subsamples contain more than d + 1 observations.
Updated hdMTD() to forward cut_fraction to hdMTD_FSC() and record it in the returned object’s settings.

Added computational-cost warnings to the documentation of hdMTD_BIC() and hdMTD_CUT(), emphasizing that unrestricted candidate sets may substantially increase running time and memory requirements.
Recommended using a reduced candidate lag set S, when possible, for example based on the output of hdMTD_FS().
Clarified in the documentation of functions that accept chain samples that missing values (NA) are not allowed.
Documented the chronological sample split and the role of cut_fraction in hdMTD_FSC().
Updated the documentation of the S3 methods for classes MTD, MTDest, and hdMTD to include explicit usage entries for the documented methods.
Updated the documentation of tempdata to identify INMET and BDMEP as the data source without including the unstable external URL that triggered connection failures in the CRAN incoming checks.

Added an official package vignette with replication materials for the examples and analyses presented in Section 5 of the article describing the hdMTD package.
Included pre-computed results for computationally intensive analyses so that the vignette can be built without rerunning long computations.

Added an automated test confirming that inference functions reject samples containing missing values with a clear error message.

Simplified the S3 class hierarchy: fitted objects of class MTDest now inherit from MTD.
Removed redundant class checks from S3 methods, relying on method dispatch.
Eliminated duplicated S3 methods (coef.MTDest, probs.MTDest), which now work via inheritance.
Streamlined plotting code: overlapping plot types in plot.MTDest are delegated to plot.MTD using NextMethod().
Simplified summary() methods by printing directly from the summary output, removing auxiliary summary-print classes.
Reorganized internal helper functions (e.g., moved PI() and sx() from utils.R to dedicated files).
Improved and consolidated user-facing documentation, explicitly listing available S3 methods and accessors in the constructors’ help pages.

Modified the tie-breaking rule in the FS (Forward Selection) procedure to ensure deterministic behavior.
Updated MTD-methods, MTDest-methods and MTD-accessors documentation to remove redundant links in the help system and streamline method listings.
Sample size is now a required argument in perfectSample().
Improved the error message in logLik.MTD() when a sample is not provided.

Added a plot.MTD() method for visualizing MTD models, including bar plots of lag contributions and mixture weights, as well as directed weighted graphs (via igraph) representing each lag-specific transition matrix.
Added a plot.MTDest() method for fitted MTDest objects, which mirrors plot.MTD() but also includes EM iteration diagnostics (log-likelihood variation per update) when available.

Accessor functions for “MTD”: pj(), p0(), lambdas(), lags(), Lambda(), states(), and transitP(). See ?MTD-accessors.
Accessor functions for “MTDest”: pj(), p0(), lambdas(), lags(), S() and states(). See ?MTD-accessors.
Accessor functions for “hdMTD”: S() and lags(). See ?MTD-accessors.
Methods for “MTD” and “MTDest” objects: added print(), summary(), coef(), logLik() and probs(). For compact inspection of lag sets, state space, mixture weights and more. See ?MTD-methods and ?MTDest-methods.
Methods for “hdMTD” objects: added print() and summary() for compact inspection of lag selection results. See ?hdMTD-methods.
Coercion: new as.MTD() to rebuild an “MTD” object from an “MTDest” fit.

probs() is now a S3 generic with methods for “MTD” and “MTDest”. Returns one-step-ahead predictive probabilities either for specific contexts (context=) or from sample rows (newdata=). If neither is supplied, it returns the full global transition matrix (transitP(object) for MTD; transitP(as.MTD(object)) for MTDest).
Renamed the sample-based estimator probs(X, S, ...) to empirical_probs(X, S, ...) to avoid ambiguity: empirical_probs() estimates transition probabilities from data, while probs() returns predictive probabilities from model/fit objects.

Replaced any(is.na(X)) with anyNA(X) in checkSample() for efficiency and clarity.

Removed unused datasets (raindata, sleepscoring, testChains).
Updated examples to use simulated data (via perfectSample()) instead of the removed testChains dataset.
Internal helpers marked @keywords internal so they no longer appear in help(package="hdMTD").