scikit Mixtypes of Y error

Question

Hi I'm a scikit newbie here. I'm trying to train the computer that given an array of float decide between the 3 classes. I was classifying the classes as 0, 0.5, and 1. I also tried 0, 1.0, and 2.0 . I still get the following error:

File "/Library/Python/2.7/site-packages/sklearn/utils/multiclass.py", line 85, in     unique_labels
raise ValueError("Mix type of y not allowed, got types %s" % ys_types)
ValueError: Mix type of y not allowed, got types set(['continuous', 'multiclass'])

I have no idea what that error means

Answer 1

Try using integer types for your target labels. Or, perhaps better, use string labels like ['a', 'b', 'c'] but with more descriptive names.

If you check the code for this file multiclass.py (code is here) and look for the function type_of_target, you'll see that it is well-documented for this case.

Because some of the data are treated as float type (when 0.5 is included), it will believe you've got continuous-valued outputs, which won't do for multiclass discrete classification.

On the other hand, it will look at [0, 1.0, 2.0] like it is one integer and two floats, which is why you get both continuous and multiclass. Switching the last example to [0, 1, 2] should work. The documentation also makes it sound like switching to [0.0, 1.0. 2.0] would also work, but be careful and test that first.