Changelog • forecasttools

forecasttools (development version)

Simplified of hubverse package installation
Added for approximating discrete category probabilities from continuous quantilies.
Addded quantiles_to_category_cdf() and quantiles_to_category_pmf() functions to approximate discrete category probabilities from quantiles for an underlying continuous distribution.
Added support For NHSN PRISM thresholds to PRISM cutpoint handling
Added vintaged PRISM rate reference populations.
Updated NSSP PRISM thresholds to 2026-27 values.

New ggplot2 helper functions and custom geometric objects: sym_limits(), geom_path_point(), geom_line_point()
New plotting function for forecast evaluation: plot_pred_obs_pointintervals()
New forecast scoring helper functions: summarize_scores_with_baseline(), get_shared_forecasts(), get_subset_forecasts(), filter_to_shared_forecasts(), filter_to_subset_forecasts().
New Hubverse to scorable table approach leveraging the hubverse “oracle output” target data format.
Refactor of data.cdc.gov data pull helper functions. Deprecation of pull_nhsn() in favor of new function pull_data_cdc_gov_dataset().

New function: modify_reference_date()
BREAKING CHANGE: Various epiweek/epiyear functions no longer default to the MMWR epiweek standard and require the user to specify a standard explicitly. To restore previous behavior, set epiweek_standard = "MMWR" explicitly.

BREAKING CHANGE: get_hubverse_table() has been removed and replaced by get_hubverse_quantile_table(), get_flusight_hub_table() and get_covid_hub_table().
horizons_from_target_end_dates is now vectorized
New function: hub_quantiles_to_median_qi()

BREAKING CHANGE: Functions to prepare hubs for scoring (hubverse_table_with_obs() and quantile_table_to_scorable()) now no longer assume that forecast tables contain a single target. This will break existing scoring workflows that use truth data without a target column. It can be addressed by setting the id_cols argument hubverse_table_with_obs() or quantile_table_to_scorable() explicitly.
New function for calculating forecast horizons from target end dates: horizons_from_target_end_dates()
New epiweek rounding functions: floor_mmwr_epiweek, ceiling_mmwr_epiweek(), floor_isoweek(), and ceiling_isoweek().
Smaller-than-memory parquet file I/O now handled via nanoparquet (previously handled via arrow, reducing forecasttools package weight and installation time.

BREAKING CHANGE: target_end_dates_from_horizons() and get_hubverse_table() now work with both daily and weekly horizons. To restore functionality of existing code, add the argument horizon_timescale = "weeks" to these functions.

Adds optional epidate annotation to daily_to_epiweekly()
Increases testing coverage
Fixes bug in pull_nhsn() so that ordering on multiple columns works as expected.