Computes the normalized mean squared error (NMSE) between predictions
and reference values.
Usage
nmse(x, ref, trainY, na.rm = FALSE)
Arguments
- x
numeric vector of predicted values
- ref
numeric vector of reference (true) values
- trainY
numeric vector used as the normalization baseline
- na.rm
logical; whether to remove incomplete cases before the
computation. Defaults to FALSE, in which case a missing value
anywhere makes the result NA.
Value
a numeric scalar containing the normalized mean squared error
Details
The normalized mean squared error is defined as:
$$
\frac{\sum (ref - x)^2}{\sum (ref - mean(trainY))^2}
$$
The denominator represents the squared deviation from the mean of the
training response, providing a baseline for comparison.
If the denominator is zero, NA is returned.
See also
mean(), sum()
Other model.metrics:
auc(),
averagePrecision(),
brierScore(),
logLoss(),
mae(),
mape(),
mse(),
nmae(),
rmse(),
smape()
Examples
x <- c(2.5, 3.0, 2.8)
ref <- c(3.0, 2.5, 3.0)
trainY <- c(2, 3, 4, 3)
nmse(x, ref, trainY)
#> [1] 2.16
