arimasel 0.2.0

Response to CRAN reviewer comments (resubmission)

• DESCRIPTION and roxygen documentation now consistently write function names with (), e.g. auto.arima().
• Functions that previously wrote directly to the console with unconditional cat()/print() calls – cp_sets(), stationarity_test(), compare_arima(), arima_diagnose(), and ts_eda() – now return proper S3 objects (cartProduct, stationarityTest, compareArima, arimaDiagnose, tsEda respectively) with dedicated print methods, so output only appears via explicit or automatic top-level printing and is suppressed by ordinary assignment. smart_arima()’s informational notes now use message() (suppressible via suppressMessages()) instead of cat().
• The vignette’s multi-panel plotting chunk now saves and restores the user’s par() settings with oldpar <- par(...); ...; par(oldpar) instead of a partial manual reset.

Bug fixes

• fitted.cartARIMA() (and plot(type = "fitted")) previously returned a non-numeric result because stats::arima() fits have no fitted.Arima method for stats::fitted() to dispatch to. Fitted values are now computed correctly as observed minus one-step-ahead residuals.
• arima_forecast() prediction intervals were inverted (lower > upper) because the two-sided normal critical value was computed from the wrong tail (qnorm((100 - level) / 200), which is negative). Fixed to use qnorm((100 + level) / 200).
• Fixed an ORCID/CITATION inconsistency between DESCRIPTION and inst/CITATION.

New features

• cart_arima() gains a seasonal argument (list(P, D, Q, period)) for exhaustive seasonal ARIMA search over every combination.
• cart_arima() and arima_forecast() gain xreg / newxreg support for regression with ARIMA errors.
• cart_arima() gains a parallel / n_cores option for multi-core candidate model fitting (via parallel::mclapply, with an automatic serial fallback on Windows).
• New arima_cv() for rolling-origin (expanding-window) cross-validation (“backtesting”) of the selected model order, with print.arimaCV, summary.arimaCV, and plot.arimaCV methods.
• New seasonal_strength() (STL-based trend/seasonal strength, Wang, Smith & Hyndman 2006) and suggest_D() (seasonal differencing recommendation).
• plot.cartARIMA() gains a type = "seasonal" STL decomposition panel.
• cp_sets() and compare_arima() are now seasonal-aware.
• New ts_features(): a compact, scale-free feature summary of a time series (trend/seasonal strength, spectral entropy, ACF(1), lumpiness, stability, ADF statistic) in the spirit of Hyndman, Wang and Laptev (2015).
• New ts_eda(): a one-call exploratory data analysis routine producing a multi-panel plot (time plot, distribution, ACF/PACF, and, for seasonal series, seasonal subseries and seasonal lag plots) plus a console summary of ts_features(), stationarity_test(), and recommended differencing orders.
• New smart_arima(): a feature-guided wrapper around cart_arima() that uses ts_features()/suggest_d()/suggest_D() to automatically narrow the differencing (and, where relevant, seasonal) search space before running the exhaustive Cartesian-product search – a lightweight, transparent alternative to black-box automatic model selection.

Other changes

• Package version bumped to 0.2.0.
• Maintainer email and DESCRIPTION/CITATION corrected/aligned.
• Expanded test suite covering the new seasonal, xreg, cross-validation, and EDA/feature-based functionality, plus regression tests for both fixed bugs.
• Vignette extended with exploratory data analysis, seasonal search, exogenous regressor, cross-validation, and feature-guided search examples.

arimasel 0.1.0

• Initial release.