Github user dbtsai commented on a diff in the pull request:
https://github.com/apache/spark/pull/10743#discussion_r49649843
--- Diff:
mllib/src/main/scala/org/apache/spark/ml/classification/LogisticRegression.scala
---
@@ -339,9 +339,11 @@ class LogisticRegression @Since("1.2.0") (
b = \log{P(1) / P(0)} = \log{count_1 / count_0}
}}}
*/
- initialCoefficientsWithIntercept.toArray(numFeatures)
- = math.log(histogram(1) / histogram(0))
- }
+ if (histogram.length >= 2) { // check to make sure indexing into
histogram(1) is safe
+ initialCoefficientsWithIntercept.toArray(numFeatures) = math.log(
+ histogram(1) / histogram(0))
--- End diff --
In this case, the whole training step can be skipped. Currently, we only
support binary LoR, so the max of `histogram.length` will be two. In LiR, when
the `yStd == 0.0`, the model will be returned immediately without training, see
https://github.com/feynmanliang/spark/blob/SPARK-12804/mllib/src/main/scala/org/apache/spark/ml/regression/LinearRegression.scala#L226
We can do similar thing here like
```scala
if (histogram.length == 2) {
if (histogram(0) == 0.0) {
model = (new LogisticRegressionModel(uid, Vectors.sparse(numFeatures,
Seq()), Double.PositiveInfinity))
return model
} else {
initialCoefficientsWithIntercept.toArray(numFeatures) = math.log(
histogram(1) / histogram(0))
} else if (histogram.length == 1) {
model = (new LogisticRegressionModel(uid, Vectors.sparse(numFeatures,
Seq()), Double.NegativeInfinity))
return model
} else {
some excpetion
}
}
```
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
