Skip to content

Fix brier_class() case weight normalization #645

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
taekop wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix brier_class() case weight normalization #645

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions NEWS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,9 @@
# yardstick (development version)

## Bug fixes

* `brier_class()` now computes the correct result with `case_weights` when the weights are on very different scales, rather than silently renormalizing them with a softmax (#633).

# yardstick 1.4.0

## Breaking Changes
Expand Down
8 changes: 0 additions & 8 deletions R/prob-brier_class.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,14 +182,6 @@ brier_ind <- function(truth, estimate, case_weights = NULL) {
resids <- resids[not_missing, , drop = FALSE]
case_weights <- case_weights[not_missing]

# Normalize weights (in case negative weights)
# subtracting max to avoid Inf in calculations
# exp(x - max(x)) / sum(exp(x - max(x)))
# = (exp(x) / exp(max(x))) / (sum(exp(x)) / exp(max(x)))
# = exp(x) / sum(exp(x))
case_weights <- case_weights - max(case_weights)
case_weights <- exp(case_weights) / sum(exp(case_weights))

res <- sum(resids * case_weights) / (2 * sum(case_weights))
res
}
Expand Down
2 changes: 1 addition & 1 deletion tests/testthat/test-orderedprob-ranked_prob_score.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,7 +52,7 @@ test_that("Case weights calculations are correct", {
hpc_cv$obs <- as.ordered(hpc_cv$obs)
hpc_cv$weights <- rep(c(0, 1), c(100, nrow(hpc_cv) - 100))
# With orf:::rps(as.matrix(hpc_cv[-(1:100), 3:6]), hpc_cv$obs[-(1:100)])
hpc_exp <- 0.0868733
hpc_exp <- 0.0875957

expect_equal(
yardstick:::ranked_prob_score_vec(
Expand Down
12 changes: 11 additions & 1 deletion tests/testthat/test-prob-brier_class.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,12 +114,22 @@ test_that("doesn't produce NaN with high valued case weights (#614)", {
wts <- rep(1, (nrow(df)))
wts_high <- wts * 5000

expect_identical(
expect_equal(
brier_class_vec(df$truth, df$Class1, case_weights = wts),
brier_class_vec(df$truth, df$Class1, case_weights = wts_high)
)
})

test_that("case weights match uncounted data with widely spread weights (#633)", {
hpc_wts <- hpc_cv
hpc_wts$wts <- rep(1:10, length.out = nrow(hpc_wts))

expect_equal(
brier_class(hpc_wts, obs, VF:L, case_weights = wts)[[".estimate"]],
brier_class(tidyr::uncount(hpc_wts, wts), obs, VF:L)[[".estimate"]]
)
})

test_that("`event_level = 'second'` works", {
df <- two_class_example

Expand Down