H2O error logistic regression

Roeieber

Hello,

I am a student using RM for my education.
I have a dataset from internet and want to use logistic regression on it but when running the model I get this message: Model training error (H2O) illegal arguments for GLM mode: ERRR of field etc etc.

Does somebody know what i did wrong?

Best regards,
Robert

varunm1

Hello @Roeieber

The error implies that you are trying to train that model on a dataset that has a single label. I know you are not really trying to do this, but the data entering into the logistic regression is like that.

Reasons:
1. You did a split validation on data that is highly imbalanced. I recommend you select stratified sampling for this after setting a label role for a column in our data.
2. Also, you made the process a bit complex by adding a set role and numerical to binomial operators after the splitting of data. This is not necessary and also has some issues while encoding numerical to binomial.

I changed the process and attached it. Import it, test it and inform if you need more information.

Hope this helps.

varunm1

Hello @Roeieber

The error implies that you are trying to train that model on a dataset that has a single label. I know you are not really trying to do this, but the data entering into the logistic regression is like that.

Reasons:
1. You did a split validation on data that is highly imbalanced. I recommend you select stratified sampling for this after setting a label role for a column in our data.
2. Also, you made the process a bit complex by adding a set role and numerical to binomial operators after the splitting of data. This is not necessary and also has some issues while encoding numerical to binomial.

I changed the process and attached it. Import it, test it and inform if you need more information.

Hope this helps.

Roeieber

Hello Varun,

Thanks for your help.
I had to take out the numerical to binomial operator, but then it worked.
I also read about the stratified split and understand why this could be better. 

Thnk you again.
Best regards,
Robert