Changelog • augury

augury 0.3.3

All predict_...() functions now treat confidence bounds in the same way as response and predicted values. Upper and lower bounds are generated in full in pred_upper_col and pred_lower_col and then the existing bounds replaced based on replace_obs and presence of non-missing values.

Use obs_filter in all predict_... functions to replace replace_filter, allowing not just filtering of when to replace observations, but also not fitting models when not being used to improve speed and reduce errors if insufficient data for certain types of modeling.
Add expand_df() function to allow easy generation of data frames with explicit missing values prior to passing to predict_... functions.

Add back extrapolation (flat) to predict_simple().
Add predict_aarr() to allow the use of AARR for forecasting prevalence data.
Implement replace_filter in all predict_... functions that allows for select use of predicted data based on number of observations so that different models can be used for different data typologies.
Change defaults for group_col and sort_col to "iso3" and "year" respectively, since they are by far the most common usage.

predict_..._avg_trend() functions implemented to allow the fitting of models by group and application of that trend to base data.
Added in R-squared and root mean change error metrics to model_error.
Weighted averaging option added to predict_average().

Added in mean absolute scaled error, median absolute error, and confidence bounds assessment to error metrics.
Refitted all functions to perform grouped modeling with the group_models argument, removing the grouped_predict_... function aliases.
Fix general functionality to support model building and testing.
Add in scale and probit arguments to predict_... functions to enable automatic scaling and transforming of response variables prior to model fitting.