InternalPerformanceSummaryFunctions: Internal summary functions for performance objects from...

In AndrewLawrence/dCVnet: doubly cross-validated elastic-net regularised generalised linear models

Internal summary functions for performance objects from different model families.

Description

For binomial and multinomial family models: Various performance metrics from confusionMatrix; AUC (or mAUC) (auc); Brier Score (brier)

For Gaussian/Poisson family models: Root Mean Square Error (RMSE; rmse); Mean Absolute Error (MAE; mae); Pearson's correlation r and r^2 between observed and predicted (R, R2); simple linear calibration intercept and slope (cal_Intercept, cal_Slope). Brier score, which in this context is the Mean Square Error == RMSE^2; RMSE and MSE scaled by the S.D. of y (SDScaledRMSE, SDScaledMAE); Optionally Somer's Dxy can be returned (SomersDxy; SomersDelta). For the Poisson family performance measures above but calculated from log1p transformed data are also returned.

For Cox family models: Harrell's C-index and Dxy using the rcorr.cens function.

For multivariate Gaussian family models: Runs perf_cont for each outcome.

Usage

perf_nomial(object, short = FALSE, somersD = FALSE, pvprevalence = "observed")

perf_cont(object, short = FALSE, somersD = FALSE, pvprevalence = "observed")

perf_cox(object, short = FALSE, somersD = FALSE, pvprevalence = "observed")

perf_mgaussian(
  object,
  short = FALSE,
  somersD = FALSE,
  pvprevalence = "observed"
)

Arguments

object	a performance object.
short	(bool) return a core set of performance measures.
somersD	(bool) calculate the computationally expensive Somers' D for certain families (gaussian, poisson)
pvprevalence	argument for adjustment of PPV/NPV calculation for binomial or multinomial families. Either "observed" to use the observed prevalence, or a number [0,1] (for binomial), or a vector of length n_categories (for multinomial) containing numerics in [0,1].

Functions

• perf_nomial(): Used for binomial and multinomial families

• perf_cont(): Used for gaussian and poisson families

• perf_cox(): Used for cox models

• perf_mgaussian(): Used for multivariate gaussian models

AndrewLawrence/dCVnet documentation built on Sept. 24, 2024, 5:24 a.m.