temp

mllib/src/main/scala/org/apache/spark/mllib/optimization/Gradient.scala

@@ -94,12 +94,24 @@ class LogisticGradient extends Gradient {
       weights: Vector,
       cumGradient: Vector): Double = {
     val margin = -1.0 * dot(data, weights)
-    val gradientMultiplier = (1.0 / (1.0 + math.exp(margin))) - label
+    /**
+     * gradientMultiplier = (1.0 / (1.0 + math.exp(margin))) - label
+     * However, the first part of gradientMultiplier can be potentially suffered from overflow,
+     * so we use the equivalent formula but more numerically stable.
+     */
+    val gradientMultiplier =
+      if (margin > 0.0) {
+        val temp = math.exp(-margin)
+        temp / (1.0 + temp) - label
+      } else {
+        1.0 / (1.0 + math.exp(margin)) - label
+      }
     axpy(gradientMultiplier, data, cumGradient)
     if (label > 0) {
-      math.log1p(math.exp(margin))
+      // The following is equivalent to log(1 + exp(margin)) but more numerically stable.
+      MLUtils.log1pExp(margin)
     } else {
-      math.log1p(math.exp(margin)) - margin
+      MLUtils.log1pExp(margin) - margin
     }
   }
 }