In USGS-R/trainR: Example package for learning how to develop R packages.

There are two root mean squared error functions. One simply calculates the rmse for given predicted and observed values:

$$ rmse = \sqrt{\frac{1}{N}\sum_{i=1}^{N}(y-\hat{y})^2} $$

Here is a working example:

require(ggplot2)
require(trainR)

# create observed
obs <- 1:50

# create error and predicted
e <- rnorm(n=50,mean=0,sd=5)
pred <- obs + e

# run function
rmse(obs,pred)

# plot
qplot(obs,pred)

There is another function that calculates the RMSE by year. Here is a working example:

# generate year and observed
year <- c(rep(1,25),rep(2,25),rep(3,25))
obs <- c(1:75)

# generate errors that change by year
e <- c(rnorm(n=25,mean=0,sd=2),rnorm(n=25,mean=0,sd=10),rnorm(n=25,mean=0,sd=50))
pred <- obs + e

# create data frame
df <- data.frame(year,obs,pred)

# run function
rmse.by.year(df)