Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[R] Rename BIAS -> (Intercept) (#10337)

Browse Source
This commit is contained in:
david-cortes 2024-05-30 13:43:32 +02:00 committed by GitHub
parent bc6c993aaa
commit 8998733ef4
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 10 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
R-package/R/xgb.Booster.R
View File

@ -249,7 +249,7 @@ xgb.get.handle <- function(object) {
#' summary(rowSums(pred_contr) - qlogis(pred)) #' summary(rowSums(pred_contr) - qlogis(pred))
#' # for the 1st record, let's inspect its features that had non-zero contribution to prediction: #' # for the 1st record, let's inspect its features that had non-zero contribution to prediction:
#' contr1 <- pred_contr[1,] #' contr1 <- pred_contr[1,]
#' contr1 <- contr1[-length(contr1)] # drop BIAS #' contr1 <- contr1[-length(contr1)] # drop intercept
#' contr1 <- contr1[contr1 != 0] # drop non-contributing features #' contr1 <- contr1[contr1 != 0] # drop non-contributing features
#' contr1 <- contr1[order(abs(contr1))] # order by contribution magnitude #' contr1 <- contr1[order(abs(contr1))] # order by contribution magnitude
#' old_mar <- par("mar") #' old_mar <- par("mar")
@ -473,7 +473,7 @@ predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FA
.Call(XGSetArrayDimInplace_R, arr, rev(shape)) .Call(XGSetArrayDimInplace_R, arr, rev(shape))
cnames <- if (!is.null(colnames(newdata))) c(colnames(newdata), "BIAS") else NULL cnames <- if (!is.null(colnames(newdata))) c(colnames(newdata), "(Intercept)") else NULL
n_groups <- shape[2] n_groups <- shape[2]
## Needed regardless of whether strict shape is being used. ## Needed regardless of whether strict shape is being used.

2
R-package/man/predict.xgb.Booster.Rd
View File

@ -211,7 +211,7 @@ str(pred_contr)
summary(rowSums(pred_contr) - qlogis(pred)) summary(rowSums(pred_contr) - qlogis(pred))
# for the 1st record, let's inspect its features that had non-zero contribution to prediction: # for the 1st record, let's inspect its features that had non-zero contribution to prediction:
contr1 <- pred_contr[1,] contr1 <- pred_contr[1,]
contr1 <- contr1[-length(contr1)] # drop BIAS contr1 <- contr1[-length(contr1)] # drop intercept
contr1 <- contr1[contr1 != 0] # drop non-contributing features contr1 <- contr1[contr1 != 0] # drop non-contributing features
contr1 <- contr1[order(abs(contr1))] # order by contribution magnitude contr1 <- contr1[order(abs(contr1))] # order by contribution magnitude
old_mar <- par("mar") old_mar <- par("mar")

10
R-package/tests/testthat/test_helpers.R
View File

@ -101,7 +101,7 @@ test_that("predict feature contributions works", {
# gbtree binary classifier # gbtree binary classifier
expect_error(pred_contr <- predict(bst.Tree, sparse_matrix, predcontrib = TRUE), regexp = NA) expect_error(pred_contr <- predict(bst.Tree, sparse_matrix, predcontrib = TRUE), regexp = NA)
expect_equal(dim(pred_contr), c(nrow(sparse_matrix), ncol(sparse_matrix) + 1)) expect_equal(dim(pred_contr), c(nrow(sparse_matrix), ncol(sparse_matrix) + 1))
expect_equal(colnames(pred_contr), c(colnames(sparse_matrix), "BIAS")) expect_equal(colnames(pred_contr), c(colnames(sparse_matrix), "(Intercept)"))
pred <- predict(bst.Tree, sparse_matrix, outputmargin = TRUE) pred <- predict(bst.Tree, sparse_matrix, outputmargin = TRUE)
expect_lt(max(abs(rowSums(pred_contr) - pred)), 1e-5) expect_lt(max(abs(rowSums(pred_contr) - pred)), 1e-5)
# must work with data that has no column names # must work with data that has no column names
@ -114,14 +114,14 @@ test_that("predict feature contributions works", {
# gbtree binary classifier (approximate method) # gbtree binary classifier (approximate method)
expect_error(pred_contr <- predict(bst.Tree, sparse_matrix, predcontrib = TRUE, approxcontrib = TRUE), regexp = NA) expect_error(pred_contr <- predict(bst.Tree, sparse_matrix, predcontrib = TRUE, approxcontrib = TRUE), regexp = NA)
expect_equal(dim(pred_contr), c(nrow(sparse_matrix), ncol(sparse_matrix) + 1)) expect_equal(dim(pred_contr), c(nrow(sparse_matrix), ncol(sparse_matrix) + 1))
expect_equal(colnames(pred_contr), c(colnames(sparse_matrix), "BIAS")) expect_equal(colnames(pred_contr), c(colnames(sparse_matrix), "(Intercept)"))
pred <- predict(bst.Tree, sparse_matrix, outputmargin = TRUE) pred <- predict(bst.Tree, sparse_matrix, outputmargin = TRUE)
expect_lt(max(abs(rowSums(pred_contr) - pred)), 1e-5) expect_lt(max(abs(rowSums(pred_contr) - pred)), 1e-5)
# gblinear binary classifier # gblinear binary classifier
expect_error(pred_contr <- predict(bst.GLM, sparse_matrix, predcontrib = TRUE), regexp = NA) expect_error(pred_contr <- predict(bst.GLM, sparse_matrix, predcontrib = TRUE), regexp = NA)
expect_equal(dim(pred_contr), c(nrow(sparse_matrix), ncol(sparse_matrix) + 1)) expect_equal(dim(pred_contr), c(nrow(sparse_matrix), ncol(sparse_matrix) + 1))
expect_equal(colnames(pred_contr), c(colnames(sparse_matrix), "BIAS")) expect_equal(colnames(pred_contr), c(colnames(sparse_matrix), "(Intercept)"))
pred <- predict(bst.GLM, sparse_matrix, outputmargin = TRUE) pred <- predict(bst.GLM, sparse_matrix, outputmargin = TRUE)
expect_lt(max(abs(rowSums(pred_contr) - pred)), 1e-5) expect_lt(max(abs(rowSums(pred_contr) - pred)), 1e-5)
# manual calculation of linear terms # manual calculation of linear terms
@ -137,7 +137,7 @@ test_that("predict feature contributions works", {
expect_is(pred_contr, "list") expect_is(pred_contr, "list")
expect_length(pred_contr, 3) expect_length(pred_contr, 3)
for (g in seq_along(pred_contr)) { for (g in seq_along(pred_contr)) {
expect_equal(colnames(pred_contr[[g]]), c(colnames(iris[, -5]), "BIAS")) expect_equal(colnames(pred_contr[[g]]), c(colnames(iris[, -5]), "(Intercept)"))
expect_lt(max(abs(rowSums(pred_contr[[g]]) - pred[, g])), 1e-5) expect_lt(max(abs(rowSums(pred_contr[[g]]) - pred[, g])), 1e-5)
} }
@ -151,7 +151,7 @@ test_that("predict feature contributions works", {
byrow = TRUE byrow = TRUE
) )
for (g in seq_along(pred_contr)) { for (g in seq_along(pred_contr)) {
expect_equal(colnames(pred_contr[[g]]), c(colnames(iris[, -5]), "BIAS")) expect_equal(colnames(pred_contr[[g]]), c(colnames(iris[, -5]), "(Intercept)"))
expect_lt(max(abs(rowSums(pred_contr[[g]]) - pred[, g])), float_tolerance) expect_lt(max(abs(rowSums(pred_contr[[g]]) - pred[, g])), float_tolerance)
# manual calculation of linear terms # manual calculation of linear terms
coefs <- c(coefs_all[-1, g], coefs_all[1, g]) # intercept needs to be the last coefs <- c(coefs_all[-1, g], coefs_all[1, g]) # intercept needs to be the last

4
R-package/tests/testthat/test_interactions.R
View File

@ -48,7 +48,7 @@ test_that("predict feature interactions works", {
intr <- predict(b, dm, predinteraction = TRUE) intr <- predict(b, dm, predinteraction = TRUE)
expect_equal(dim(intr), c(N, P + 1, P + 1)) expect_equal(dim(intr), c(N, P + 1, P + 1))
# check assigned colnames # check assigned colnames
cn <- c(letters[1:P], "BIAS") cn <- c(letters[1:P], "(Intercept)")
expect_equal(dimnames(intr), list(NULL, cn, cn)) expect_equal(dimnames(intr), list(NULL, cn, cn))
# check the symmetry # check the symmetry
@ -60,7 +60,7 @@ test_that("predict feature interactions works", {
# diagonal terms for features 3,4,5 must be close to zero # diagonal terms for features 3,4,5 must be close to zero
expect_lt(Reduce(max, sapply(3:P, function(i) max(abs(intr[, i, i])))), 0.05) expect_lt(Reduce(max, sapply(3:P, function(i) max(abs(intr[, i, i])))), 0.05)
# BIAS must have no interactions # Intercept must have no interactions
expect_lt(max(abs(intr[, 1:P, P + 1])), 0.00001) expect_lt(max(abs(intr[, 1:P, P + 1])), 0.00001)
# interactions other than 2 x 3 must be close to zero # interactions other than 2 x 3 must be close to zero