RemixAutoML (version 0.4.2)

ParDepCalPlots: ParDepCalPlots automatically builds partial dependence calibration plots for model evaluation

Description

This function automatically builds partial dependence calibration plots and partial dependence calibration boxplots for model evaluation using regression, quantile regression, and binary and multinomial classification

Usage

ParDepCalPlots(
  data,
  PredictionColName = c("PredictedValues"),
  TargetColName = c("ActualValues"),
  IndepVar = c("Independent_Variable_Name"),
  GraphType = c("calibration"),
  PercentileBucket = 0.05,
  FactLevels = 10,
  Function = function(x) mean(x, na.rm = TRUE)
)

Arguments

data

Data containing predicted values and actual values for comparison

PredictionColName

Predicted values column names

TargetColName

Target value column names

IndepVar

Independent variable column names

GraphType

calibration or boxplot - calibration aggregated data based on summary statistic; boxplot shows variation

PercentileBucket

Number of buckets to partition the space on (0,1) for evaluation

FactLevels

The number of levels to show on the chart (1. Levels are chosen based on frequency; 2. all other levels grouped and labeled as "Other")

Function

Supply the function you wish to use for aggregation.

Value

Partial dependence calibration plot or boxplot

See Also

Other Model Evaluation and Interpretation: AutoLimeAid(), EvalPlot(), LimeModel(), RedYellowGreen(), threshOptim()

Examples

# NOT RUN {
# Create fake data
data <- RemixAutoML::FakeDataGenerator(
  Correlation = 0.70, N = 10000000, Classification = FALSE)
data.table::setnames(data, "Independent_Variable2", "Predict")

# Build plot
Plot <- RemixAutoML::ParDepCalPlots(
  data,
  PredictionColName = "Predict",
  TargetColName = "Adrian",
  IndepVar = "Independent_Variable1",
  GraphType = "calibration",
  PercentileBucket = 0.20,
  FactLevels = 10,
  Function = function(x) mean(x, na.rm = TRUE))
# }