Using the R dataset diamonds, I was wondering how the initial_split function stratifies the continuous variable "x" in order to a create test and training set? ie initial_split(diamonds, probability=0.8, strata=x). I had read that the continuous data "x" is separated sequentially into bins, but I couldn't find out how many bins, or if the number of bins could be set by the user?
How does tidymodels initial_split function deal with a strata variable that is continuous?
Viewed 546
1 Answers
The number of breaks used by initial_split() when stratifying against a continuous variable is specified by the breaks argument. breaks defaults to 4 bins.
library(rsample)
library(ggplot2)
data("diamonds")
initial_split(diamonds, strata = x, breaks = 4)
#> <Analysis/Assess/Total>
#> <40453/13487/53940>
What is happening under the hood is that make_strata() is used to create a factor variable that is then for stratification. The factor variable will be created based on percentiles to ensure that the strata are of roughly uniform size.
table(make_strata(diamonds$x, breaks = 4))
#>
#> [0,4.71] (4.71,5.7] (5.7,6.54] (6.54,10.7]
#> 13521 13665 13379 13375
table(make_strata(diamonds$x, breaks = 10))
#>
#> [0,4.36] (4.36,4.54] (4.54,4.82] (4.82,5.23] (5.23,5.7] (5.7,6.08]
#> 5774 5032 5394 5463 5523 5186
#> (6.08,6.42] (6.42,6.7] (6.7,7.31] (7.31,10.7]
#> 5407 5429 5458 5274
Related