Github user dbtsai commented on a diff in the pull request:
https://github.com/apache/spark/pull/6386#discussion_r30958988
--- Diff:
mllib/src/main/scala/org/apache/spark/mllib/classification/LogisticRegression.scala
---
@@ -363,4 +370,34 @@ class LogisticRegressionWithLBFGS
new LogisticRegressionModel(weights, intercept, numFeatures,
numOfLinearPredictor + 1)
}
}
+
+ /**
+ * Run the algorithm with the configured parameters on an input RDD
+ * of LabeledPoint entries starting from the initial weights provided.
+ * If a known updater is used calls the ml implementation, to avoid
+ * applying a regularization penalty to the intercept, otherwise
+ * defaults to the mllib implementation. If more than two classes
+ * or feature scaling is disabled, always uses mllib implementation.
+ */
+ override def run(input: RDD[LabeledPoint], initialWeights: Vector):
LogisticRegressionModel = {
+ // ml's Logisitic regression only supports binary classifcation
currently.
+ if (numOfLinearPredictor == 1 && useFeatureScaling) {
+ def runWithMlLogisitcRegression(elasticNetParam: Double) = {
--- End diff --
This will be another feature.
When people train LoR/LiR with multiple lambda of regularizations for
cross-validation, the training algorithm will start from the largest lambda and
return the model. The model will be used as the initial condition for the
second largest lambda. The process will be repeated until all the lambdas are
trained. By using the previous model as initial weights, the convergence rate
will be a way faster. http://www.jstatsoft.org/v33/i01/paper
As a result, in order to do so, we need to have ability to specify initial
weights. Feel free to add private API to set weights. If the dim of weights is
different from the data, then we can use the default one as initial condition.
PS, once this private api is added, we can hook it up with CrossValidation
API to train multiple lambdas efficiently. Currently, with multiple lambda, we
train from scratch without using the information from previous results. No JIRA
now, you can open one if you are interested in this.
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
