Skip to content

[SPARK-5099][Mllib] Simplify logistic loss function #3899

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

Closed
viirya wants to merge 5 commits into from
Closed

[SPARK-5099][Mllib] Simplify logistic loss function #3899

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
2bc5712
Simplify loss function.
viirya Jan 5, 2015
a3f83ca
Fix a bug.
viirya Jan 6, 2015
72a295e
Revert LogLoss back and add more considerations in Logistic Loss.
viirya Jan 6, 2015
0aa51e4
Further simplified.
viirya Jan 6, 2015
91a3860
Modified for comment.
viirya Jan 7, 2015
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
12 changes: 9 additions & 3 deletions mllib/src/main/scala/org/apache/spark/mllib/optimization/Gradient.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,11 +64,17 @@ class LogisticGradient extends Gradient {
val gradientMultiplier = (1.0 / (1.0 + math.exp(margin))) - label
val gradient = data.copy
scal(gradientMultiplier, gradient)
val minusYP = if (label > 0) margin else -margin

// log1p is log(1+p) but more accurate for small p
// Following two equations are the same analytically but not numerically, e.g.,
// math.log1p(math.exp(1000)) == Infinity
// 1000 + math.log1p(math.exp(-1000)) == 1000.0
val loss =
if (label > 0) {
math.log1p(math.exp(margin)) // log1p is log(1+p) but more accurate for small p
if (minusYP < 0) {
math.log1p(math.exp(minusYP))
} else {
math.log1p(math.exp(margin)) - margin
math.log1p(math.exp(-minusYP)) + minusYP
}

(gradient, loss)
Expand Down