Changes in version 0.1.4 (2025-12-18)                  

Fixes and improvements

  - 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.

                 Changes in version 0.1.3 (2025-11-01)                  

Fixes

  - 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.

New Features

  - 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.

                 Changes in version 0.1.2 (2025-09-28)                  

New

  - 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.

Changes

  - 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.

Fixes

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

Package cleanup

  - 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").

                 Changes in version 0.1.1 (2025-08-27)                  

  - Relicensed the package from MIT to GPL-3.
  - Removed an unintended README.md file from the package source.