resiPlot: resiPlot

In SixiangHu/DataMan: Data Summary, One-way Interactive Plot and Model Visualisation Analysis

Description

This function assess the residual using given actual and predicted values.

Usage

resiPlot(act,pred,weight=NULL,exposure=NULL,bucket=20)

Arguments

act	numerical vector for actual observation.
pred	numerical vector for model preidctions. It must have the same length as act.
weight	numerical vector to give weight to observations.
exposure	numerical vector to give exposure to observations.
bucket	Integer. It specifies the number of bucket of the AvsE plot.
residualFun	Character. Name of residual function. The first parameter must be actual, and second parameter must be predictions. By default, residual function is difference between 'act' and 'pred'.
...	other parameters for 'residualFun' specified.

Details

In previous version of this function, the residual in this function is defined as: Residual = actual - predicted. 'resi' or 'reisdual' function from 'stats' package are not used because this function will be used for much wider model assess (e.g. 'randomFoest', 'gbm'), and 2 functions mentioned above can only applied to 'glm' and 'lm'.

This function will give 2 plots: AvsE plot: The average actual and predicted value for each bucket, with a diagnal line for comparison. Residual vs Prediction: This plot is used to assess the baise and heterogeneity

Author(s)

Sixiang Hu

Examples

set.seed(1L)
act <- rgamma(10000,1)
pred <- act + rnorm(10000)
signDeviance <- function(y,mu){
  sign(y-mu)*sqrt(residualDeviance(y,mu,family = "gamma"))
}
resiPlot(act,pred,residualFun = "signDeviance")

SixiangHu/DataMan documentation built on May 9, 2019, 1:48 p.m.