Logistic regression says only one label

Goryeo

I'm having a hard time finding a work around with this problem. So, I used Discretize by User Specification to categorized the values into five categories (Above Average, MTM, AM, LM, and VLM) then I set its role to label. I then used Cross Validation. This is where I started encountering problems, in the training area I used a logistic regression model but apparently it cannot handle a polynomial label. After browsing the internet, I used the Polynominal by Binominal Classification model then put the Logistic Regression model there. The error was gone but when I ran the process it says that it does have sufficient capabilities in handling an example set with only one label.

Correct me if I'm wrong, as I know, Logistic regression will run as long there are two values in a label (for example like yes or no). But why is it saying that my label has only one value when in fact it has five. 

lionelderkrikor

Hi @Goryeo,

I think that due to "bad luck", one of your folds during the Cross Validation contains only examples with one value of your label
Maybe have you got an imbalanced dataset too ?
Try to use Split Validation instead Cross Validation and tell us if this error is still present...

Regards,

Lionel

Telcontar120

If you want to use Logistic Regression then you probably should copy your existing label, then use Map to take your 5 values now to turn them into 2 new values, and then turn that version into your label.  It will make subsequent processing much easier.

jacobcybulski

Also make sure you eliminate missing values which get interpreted by some models as a separate value. 

jacobcybulski

By the way, if you want to have polynomial label, why not using a different model type, such as a decision tree? If it has to be a linear model, you can also use a General Linear Model with multinomial family type.