Error running multinomial regression on a dataset with a field having constant value

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Error running multinomial regression on a dataset with a field having constant value

kundan kumar
I am running the sample multinomial regression code given in spark docs (Version 2.2.0)


LogisticRegression lr = new LogisticRegression().setMaxIter(100).setRegParam(0.3).setElasticNetParam(0.8);
LogisticRegressionModel lrModel = lr.fit(training);

But in the dataset I am adding a constant field where all the values are same.

Now, I get an error saying

2018-03-11 15:42:58,835 [main] ERROR OWLQN  - Failure! Resetting history: breeze.optimize.NaNHistory: 
2018-03-11 15:42:58,922 [main] INFO  OWLQN  - Step Size: 1.000
2018-03-11 15:42:58,938 [main] INFO  OWLQN  - Val and Grad Norm: NaN (rel: NaN) NaN
2018-03-11 15:42:58,940 [main] INFO  OWLQN  - Converged because max iterations reached


Without the constant field in the dataset everything works fine.

Please help me understand what is the reason behind this error. When I run a binary logistic regression code it runs fine even if there are constant values in a field. 

Do I really need to get rod of constant field from my dataset while running multinomial regression.

Is it a bug or this is expected ??


Thanks !!
Kundan