I checked the labels across the entire dataset and it looks like it has -1
and 1 (not the 0 and 1 I originally expected). I will try replacing the -1
with 0 and run it again.
On Mon, Mar 16, 2015 at 12:51 AM, Rishi Yadav ri...@infoobjects.com wrote:
ca you share some sample data
On Sun, Mar 15, 2015 at 8:51 PM, Rohit U rjupadhy...@gmail.com wrote:
Hi,
I am trying to run LogisticRegressionWithSGD on RDD of LabeledPoints
loaded using loadLibSVMFile:
val logistic: RDD[LabeledPoint] = MLUtils.loadLibSVMFile(sc,
s3n://logistic-regression/epsilon_normalized)
val model = LogisticRegressionWithSGD.train(logistic, 100)
It gives an input validation error after about 10 minutes:
org.apache.spark.SparkException: Input validation failed.
at
org.apache.spark.mllib.regression.GeneralizedLinearAlgorithm.run(GeneralizedLinearAlgorithm.scala:162)
at
org.apache.spark.mllib.regression.GeneralizedLinearAlgorithm.run(GeneralizedLinearAlgorithm.scala:146)
at
org.apache.spark.mllib.classification.LogisticRegressionWithSGD$.train(LogisticRegression.scala:157)
at
org.apache.spark.mllib.classification.LogisticRegressionWithSGD$.train(LogisticRegression.scala:192)
From reading this bug report (
https://issues.apache.org/jira/browse/SPARK-2575) since I am loading
LibSVM format file there should be only 0/1 in the dataset and should not
be facing the issue in the bug report. Is there something else I'm missing
here?
Thanks!