Cost functions

This method uses algebraic method assuming normal distribution of the residuals. This is done by using sd rather than RSE from lm model.

cost_MBE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_MAE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_MSE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_RMSE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_SSE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_MinErr(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_MaxErr(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_MaxAbsErr(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_MaxSqErr(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_R_squared(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_SESOItoRMSE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_PPER(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_MHE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

cost_RMHE(
  observed,
  predicted,
  SESOI_lower = 0,
  SESOI_upper = 0,
  negative_weight = 1,
  positive_weight = 1,
  na.rm = FALSE
)

Arguments

observed	Numeric vector
predicted	Numeric vector
SESOI_lower	Lower smallest effect size of interest threshold
SESOI_upper	Upper smallest effect size of interest threshold
negative_weight	How should negative residuals be weighted? Default is 1
positive_weight	How should positive residuals be weighted? Default is 1
na.rm	Should NAs be removed? Default is FALSE

Examples

data("yoyo_mas_data")

model <- lm(MAS ~ YoYoIR1, yoyo_mas_data)

observed <- yoyo_mas_data$MAS
predicted <- predict(model)

SESOI_lower <- -0.5
SESOI_upper <- 0.5

# Mean Squared Error
cost_MSE(
  observed = observed,
  predicted = predicted,
  SESOI_lower = SESOI_lower,
  SESOI_upper = SESOI_upper
)
#> [1] 0.04465622

# Mean Absolute Error
cost_MAE(
  observed = observed,
  predicted = predicted,
  SESOI_lower = SESOI_lower,
  SESOI_upper = SESOI_upper
)
#> [1] 0.1749008

# Root Mean Squared Error
cost_RMSE(
  observed = observed,
  predicted = predicted,
  SESOI_lower = SESOI_lower,
  SESOI_upper = SESOI_upper
)
#> [1] 0.2113202

# Bias
cost_MBE(
  observed = observed,
  predicted = predicted,
  SESOI_lower = SESOI_lower,
  SESOI_upper = SESOI_upper
)
#> [1] -5.921193e-15

# Sum of Squared Errors
cost_SSE(
  observed = observed,
  predicted = predicted,
  SESOI_lower = SESOI_lower,
  SESOI_upper = SESOI_upper
)
#> [1] 1.339687

# Proportion of Practically Equivalent Residuals
cost_PPER(
  observed = observed,
  predicted = predicted,
  SESOI_lower = SESOI_lower,
  SESOI_upper = SESOI_upper
)
#> [1] 0.9728044