This package contains functions that allow to quickly calculate measures of model prediction accuracy (e.g. rmse), create scatterplots that contain text informing on the agreement between the two plotted variables.
install.packages("devtools")
devtools::install_github("ptompalski/scatter")
library(scatter)
Function calc.error() is useful for quickly calculating bias and RMSE (and other) between observed and predicted values. This function also check the statistical significance of the differences. For example:
ref <- iris$Sepal.Length
est <- predict(lm(data=iris,iris$Sepal.Length~iris$Petal.Width))
calc.error(reference = ref, estimate = est)
Optionally a grouping variable can be added:
grouping_var <- iris$Species
calc.error(reference = ref, estimate = est,by = grouping_var)
I recommend setting the parameter noinfo = F during the first run and see how the absolute and relative bias are calculated.
There are two plot templates: one for scatterplot (scatter()) and one for histogram (h()).
Using:
ref <- iris$Sepal.Length
est <- predict(lm(data=iris,iris$Sepal.Length ~ iris$Petal.Width))
scatter(x = ref, y = est)
will produce a scatter plot with additional information (text box) on the bias and RMSE. Optionally you can disable this by setting info = F.
The default behaviour of h() is to plot a histogram with chosen descriptive statistics added in a corner. You can turn them off by setting info = F. You can also very easly define the width of the histogram bins (much easier than with standard hist()) with:
h(ref,0.1)
Both scatter() and h() use ggplot2 package to produce plots. The returned ggplot objects can be
further modified with additional options:
h(iris$Sepal.Length) + theme_minimal()