Weighted eCDF #5119
Weighted eCDF #5119
Conversation
|
I have no experience with ecdfs so I can't really comment on the correctness of the weighted implementation. @clauswilke or @yutannihilation do you feel confident in reviewing this? |
|
I too am not familiar with eCDF, sorry... |
|
The calculation looks correct on first glance. I might want to read it through a little more carefully before signing off on it, but fundamentally it's very simple. An eCDF is simply a cumulative sum of the ordered values, divided by the total sum. To make this weighted, you multiply each value by a weight before you sum. I have one concern though: I don't particularly like using a built-in function to calculate eCDF and a custom function to calculate weCDF. Why not use the same function for both and set the weights to 1 if not provided? |
Fair point, I don't think there is a particular reason I did it this way. Setting the weights to 1 should indeed give identical output. |
Merge branch 'main' into weighted_ecdf # Conflicts: # R/stat-ecdf.R # tests/testthat/test-stat-ecdf.R
Merge branch 'main' into weighted_ecdf # Conflicts: # R/stat-ecdf.R # tests/testthat/test-stat-ecdf.R
Merge branch 'weighted_ecdf' of https://github.com/teunbrand/ggplot2 into weighted_ecdf # Conflicts: # R/stat-ecdf.R
R/stat-ecdf.R
Outdated
| if (is.null(data$weight)) { | ||
| data_ecdf <- ecdf(data$x)(x) | ||
| } else { | ||
| data_ecdf <- wecdf(data$x, data$weight)(x) | ||
| } |
There was a problem hiding this comment.
We should just use wecdf() as per the discussion
Merge branch 'main' into weighted_ecdf # Conflicts: # R/stat-ecdf.R
This PR aims to fix #5058.
Briefly, it adds an optional
weightaesthetic tostat_ecdf(). If a weight is present, we calculate the ecdf in a different way, wherein each observations is weighted by the amount of the observation's weight relative to the sum of all weights in the group.