The hardware and bandwidth for this mirror is donated by METANET, the Webhosting and Full Service-Cloud Provider.
If you wish to report a bug, or if you are interested in having us mirror your free-software or open-source project, please feel free to contact us at mirror[@]metanet.ch.

conformalForecast 0.2.0

• Settings previously stored as model$alpha, model$symmetric, model$integrate, model$scorecast, model$lr, model$Csat and model$KI are now collected in model$args.
• forecast moved from Imports to Depends, so it is attached together with conformalForecast.
• acmcp() now accepts ma_method = "CSS" for faster MA scorecaster fitting at longer forecast horizons, while retaining "CSS-ML" as the default, and reports successful scorecasts in scorecast_times.
• New unit-test suite covering all four conformal methods, cvforecast(), update(), and the evaluation functions.

Bug fixes

• Corrected lower-tail calibration scores in scp() and acp(), ensuring that the infinite finite-sample correction is retained with the correct sign.
• kess = TRUE now requires a continuous quantile estimator (quantiletype = 4:9); an informative error is returned for discontinuous quantile estimators.
• acmcp() no longer fits MA models before its recursive scorecaster inputs are available.
• update() now works on pid(), acmcp() and acp(symmetric = TRUE) objects, and resumes from the last computed step instead of recomputing the whole history.
• update() now replays stored arguments, preserves forecasts at all frequencies, validates new data and external regressors, and handles failed final model fits. External regressors are retained by pid() and acmcp(), enabling external-regressor updates for all four conformal methods.
• pid() and acmcp() can be called with their documented defaults.
• print() and summary() no longer repeat the cross-validation header.
• cvforecast() no longer warns about arguments legitimately passed to forecastfun, and survives a failing final model fit.
• cp_times is reported per forecast horizon.
• coverage() and width() accept x, LOWER and UPPER through ..., and require a single level.
• lagmatrix() handles lags larger than the number of rows.
• Clearer error from scp() and acp() when ncal is too large.

Documentation

• The pkgdown site renders mathematics again (math-rendering: mathjax).
• Corrected the documented return classes, the Winkler measure name, the lagmatrix() error message, and several typos.
• Expanded the vignette with the conformal() interface, incremental updates, external regressors, clearer method descriptions, and updated ggplot2 calls.
• New vignette("update"), on extending a fitted conformal pipeline with newly arrived observations instead of refitting it from scratch.
• The introductory vignette now documents conformal(), and describes each method, its key formula and its tuning arguments in more detail.
• The pkgdown site indexes both vignettes under “Articles”.
• Help examples now cover raw inputs to coverage() and width() and cpforecast accuracy, and use fixed random seeds for reproducibility.

conformalForecast 0.1.1

• Changed accuracy.default to use S3 methods

conformalForecast 0.1.0

• First release.

Initial features

• Provides implementations of the main conformal prediction methods:
  • scp (split conformal prediction)
  • acp (adaptive conformal prediction)
  • pid (conformal PID control method)
  • acmcp (autocorrelated multi-step conformal prediction).
• Includes utility functions to compute coverage rates and interval widths for evaluating predictive performance.

These binaries (installable software) and packages are in development.
They may not be fully stable and should be used with caution. We make no claims about them.