NEWS.md
This update removes truth data functions that are no longer relevant. ### Updates - Remove download_raw_usafacts and preprocess_usafacts because the last release deprecated USAFacts as a truth source. - Remove preprocess_truth_for_zoltar and save_truth_for_zoltar, as the truth data being sent to Zoltar (JHU CSSE cases and deaths) stopped updating in March 2023.
This update has minor changes. ### Updates - Update download_raw_nytimes() to load county level data from separate files for each year on nytimes github repository. - Remove "USAFacts" as truth_source from load_truth(), plot_forecasts() and vignettes. - Update load_truth() to work with files stored with Git LFS in the US forecast hub. - Performance improvements in load_forecasts_repo() - Bug fix in use of && in align_forecasts, only relevant for R version 4.3.0 and higher.
This is a release focusing on new features and bug fixes in some major functions.
covidHubUtils now works with new version of scoringutils pacakge on GitHub.
All locations parameters now take location names. This parameter should be a vector of strings of fips code or CBSA codes or location names, such as “Hampshire County, MA”, “Alabama”, “United Kingdom”.
Update load_forecasts()
Support loading data from local zoltar module in load_forecasts() when source = "local_zoltar". Please follow instructions in load_forecasts_local_zoltar() to set up required environment for this functionality.
Fix bug resulting in empty ouput in load_forecats_zoltar() when verbose is TRUE.
Update warning messages and error messages in load_forecasts_zoltar().
Sort models parameter in load_forecasts() and load_latest_forecasts() so that the resulting data frame is locale-independent.
Update score_forecasts()
Rename sharpness score as dispersion.
Update all functionality to handle new version of scoringutils.
Update load_truth()
"inc case" and "inc death" target variables for "ECDC" truth source.Update get_model_metadata()
Remove as_of functionality in get_model_metadata() and get_model_designations(). as_of parameter is not available now.
Fix bug resulting in error in get_model_metadata() when a metadata file has NULL fields.
Update get_all_models()
hub parameter. It does not support loading model names for "ECDC" hub from remote hub repo for now.Update calc_target_end_date().
temporal_resolution
This is a release focusing on new features in most of the major functions.
covidHubUtils now requires additional R packages doParallel, parallel and foreach.
Update load_forecasts() and deprecate load_latest_forecasts().
The new implementation of load_forecasts() combines the functionality of the previous version of load_forecasts() and load_latest_forecasts(). Details about this change are in load_forecasts() documentation.
Rename forecast_dates to dates.
Add date_window_size to specify the number of days across each date in dates parameter to look for the most recent forecasts.
Use local data objects to validate targets parameter when source = "local_hub_repo".
Drop rows with NULLs in value column in forecast files when source = "local_hub_repo".
Add helper functiondate_to_datetime() that converts a date to a date time in the corresponding timezone based on hub and returns that date time in UTC timezone. This function is used when the user is using as_of parameter to load forecasts from zoltar only.
Add helper function reformat_forecasts() to format dataframe returned by a zoltar query.
Update plot_forecasts()
Add hub parameter to plot forecasts from US and ECDC hub.
Update validation for locations, truth_source and target_variable parameters.
Add top_layer parameter to switch layers of forecasts and truth data.
Update load_truth()
Support multiple target variables and has a new set of default values for target_variable and truth_source based on hub parameter. There are special cases for weekly aggregations when "inc hosp" is in target_variable. Please refer to the detail tab in function documentation.
Support loading truth data from covidData. as_of parameter is only supported when data_location = "covidData". Otherwise, this function will return warnings.
Add daily and weekly incident hospitalization data from ECDC source in load_truth().
Refactor code and add helper functions load_from_hub_repo(), load_from_coviddata() and aggregate_to_weekly()
Update score_forecasts()
Return true_value in function output.
Calculate one-sided quantile coverage denoted quantile_coverage0.xx.
Add metrics parameter which is a character vector of the metrics to be returned with options “abs_error”, “wis”, “wis_components”,“interval_coverage”, and “quantile_coverage”
Update save_truth_for_zoltar() to return differences between the new version and the current version of truth on zoltar server.
Update get_model_designations() to handle spaces in hub_repo_path parameters.
Add get_model_metadata() based off of get_model_designations() but to retrieve all fields in metadata.
Add preprocess_visualization_truth() to generate JSON truth file for covid19 hub visualization, and its corresponding unit tests
Add calc_cramers_dist_equal_space(), calc_cramers_dist_equal_space(), and calc_cramers_dist_one_model_pair() to calculate forecast similarities based on the approximation of Cramer’s distance.
This is a release for renaming plot_forecast() to plot_forecasts(). plot_forecast() is still available to use but will return deprecation warnings to the user.
This is a release focusing on updates that provide better interface with Zoltar and European COVID-19 Forecast Hub. The release also contains new feature updates and bug fixes in other util functions.
Update load_forecasts() and load_latest_forecasts()
Add hub parameter to specify the forecast hub for which data should be loaded.
Add as_of parameter to improve interface with the Zoltar query.
Add verbose parameter to specify whether to print out diagnostic messages.
Support source = local_hub_repo in load_forecasts(). However, loading versioned forecast files is only available through zoltar.
All models inputed into load_latest_forecasts() must be available in the selected source.
Refactor to improve efficiency.
Update plot_forecast()
Load "inc hosp" truth data from remote hub repository. The user does not need to provide truth_data parameter to plot daily incident hospitalization forecasts.
target_variable is now optional when forecast_data only has one target variable.
Add a new parameter use_median_as_point that defaults to FALSE. “TRUE” uses the median quantile and “FALSE” uses the point forecasts.
The function now errors when trying to plot multiple locations without a facet formula.
Fix bug that model legend is missing when the user is only plotting quantile forecasts.
Update quantile forecast color so that color transparency will not be overwritten by fill_transparency when plotting more than five models.
Update load_truth()
Add hub parameter to specify the forecast hub for which data should be loaded.
Add "inc hosp" target variable and "HealthData" source.
score_forecasts() has new parameter use_median_as_point that defaults to FALSE. “TRUE” uses the median quantile when calculating absolute error and “FALSE” uses the point forecasts for absolute error.
Add optional as_of parameter in get_model_designations(). Currently only support versioned model designation in local hub repo.
This is a release focusing on new scoring function and truth-processing functions. The release also contains new feature updates and bug fixes in other util functions.
covidHubUtils now requires the scoringutils package version to be at least 0.1.5.
score_forecasts() is now implemented for quantile-format forecasts to compute absolute error, weighted interval score, sharpness, overprediction, underprediction, and prediction interval coverage at any specified quantile. Minimally one should have the forecasts dataframe produced by load_forecasts() and the truth dataframe produced by load_truth() to calculate scores. If one desires to specify a subset of all available scores, one should consult this reference for valid scores in the desired_score_types vector.
preprocess_truth_for_zoltar() and save_truth_for_zoltar() are now implemented to create standard cumulative and incident death truth csv files for Zoltar.
preprocess_hospitalization() is now implemented to create standard cumulative and incident hospitalization truth csv files.
Update load_forecasts() and load_latest_forecasts()
Update default value of forecast_date_window_size to 0 inload_latest_forecasts() so that it looks for forecasts on the latest_forecast_date only.
Refactor load_latest_forecasts_repo(), splitting out functionality for reading in forecasts into a new exported function load_forecast_files_repo() that loads specific forecast files.
Standardize data format and columns types of the output.
Fix validation bug for forecast_dates when loading forecasts from zoltar. Loading functions will throw an error if all dates in forecast_dates are invalid forecast dates in Zoltar.
Update plot_forecast() to use more user-friendly color palettes when plotting a small number of intervals.
Update get_model_designations() to return NA when model designations for outdated models are not available on Zoltar.
This is a release focusing on new features in scoring functions and plotting functions.
Update plot_forecast()
Set truth_source to be optional when the user provides truth_data. However, it is still needed when show_caption = TRUE.
Remove format validation for model column in user-provided truth_data.
Support daily hospitalization plot. When target_variable = "inc hosp", the user needs to provided truth_data. Otherwise, an error will be thrown.
Add facet_nrow, facet_ncol, fill_transparency, title and subtitle.
Update get_plot_forecast_data()
Remove format validation for model column in user-provided truth_data.
When target_variable = "inc hosp", the user needs to provided truth_data. Otherwise, an error will be thrown.
This is a release focusing on new features in plotting functions.
plot_forecast() now supports faceted plots of multiple models, locations and forecast dates for one target variable.
In plot_forecast(), facet and facet_scales are equivalent to facets and scales in ggplot2::facet_wrap(). facet takes facet formula, for example facet = ~ model. facet_scales are expecting the same values for scales in ggplot2::facet_wrap(), such as "fixed", "free_y", "free_x" or "free".
If fill_by_model = TRUE, each model will be represented by a unique color. If fill_by model = FALSE, all models and selected prediction intervals will be represented by blue colors.
For simplicity, prediction interval legends will be grey in faceted plots. Morever, when the user selects more than 5 models, only 95% predicition interval is included. Otherwise, all selected prediction intervals will be plotted.
plot_forecast().