ecap: Estimate parameters for the ECAP adjustment

In ecap: Excess Certainty Adjusted Probability Estimate

Description

This function estimates 3 parameters that are needed to preform the ECAP adjustment. As such, it is meant to be used together with the predict.ecap() function included in the "ecap" package. The parameters estimated are the level of corruption and bias in the given unadjusted probability estimates, along with a tuning parameter needed to adjust the level of smoothness in the ECAP estimation.

Usage

ecap(
  unadjusted_prob,
  win_var,
  win_id,
  bias_indicator = F,
  lambda_grid = 10^seq(-6, 0, by = 0.5),
  gamma_grid = seq(0.001, 0.05, by = 0.001),
  theta_grid = seq(-4, 2, 0.1)
)

Arguments

unadjusted_prob	Numeric vector of probability estimates that you want estimate the ECAP parameters from.
win_var	A binary vector of wins and losses that correspond to the probabilities in the unadjusted_prob vector
win_id	A value that denotes a "win" (or if the event occurred) in the win_var vector.
bias_indicator	Set this equal to F if you don't want to consider bias in your estimation. Set it equal to T if you do.
lambda_grid	This is already predefined. However, you can adjust the grid of tuning parameters lambda that ECAP searches over if needed.
gamma_grid	This is already predefined. However, you can adjust the grid of gamma that ECAP searches over if needed.
theta_grid	This is already predefined. However, you can adjust the grid of theta that ECAP searches over if needed.

Value

An ecap object that can be used to adjust new probability estimates. It contains all of the tuning parameters needed to calibrate ECAP as well as diagnostic information on the estimate of g. The probabilities used to calibrate ECAP have also been ECAP corrected and are given as part of the output.

Author(s)

Bradley Rava, Peter Radchenko and Gareth M. James.

References

http://faculty.marshall.usc.edu/gareth-james/Research/Probs.pdf

Examples

set.seed(1)
p_obs <- runif(1000, 0, 1)
win_var <- rbinom(length(p_obs), 1, p_obs)
ecap_fit <- ecap(unadjusted_prob = p_obs, win_var = win_var, win_id = 1, bias_indicator = FALSE)

ecap documentation built on July 23, 2020, 9:07 a.m.