ddimmery/tidyhte: v0.2.1 Pre-release

tidyhte 0.2.1

• Bug fixes and fixes for build errors.

• Add methods to configs which allow the addition of models and moderators after the initial instantiation. This is a first step towards the eventual recipe API.

• If no moderators are specified in call to estimate_QoI, then all moderators listed in the MCATE definition are used.

• First version of the recipe API is available for use!

• Move joint effect model config to a single unified location in HTE_cfg.

• Save the names of features used in SuperLearner ensembles so that when out-of-sample predictions are requested, columns of all zeros can be imputed for features that do not exist. This imputation will simply eliminate errors when there is a factor variable with a level observed in training but not in testing. This error would never appear in the case of a continuous covariate, because it will always exist across splits, so it only eliminates errors due to one-hot-encoding factor variables.

• Add an initial draft of a vignette on methodological details undergirding the package.

• Adds a linear-only version of variable importance that calculates the reduction in residual MSE from including a particular moderator.

• Adds a Regression ROC Curve diagnostic for regression models.

• Allows access to predictions from effect regression.