Does catboost support one-hot encoding?

Question

I have labels that are one-hot encoded. I would like to use them to train and predict with a catboost classifier. However, it is giving me an error when I am fitting, saying that multiple integer values are not allowed per row for the labels. So does catboost not allow one-hot encoding for the labels? If not, how can I get catboost to work?

What parameters did you pass to `catboost`? Did you include `one_hot_max_size`? — Bob Dalgleish, Aug 11 '17 at 23:18
@BobDalgleish yep... I just passed in learning_rate, depth, iterations, and one_hot_max_size .... I did find a workaround though... — TanMath, Aug 11 '17 at 23:34

score 0 · Answer 1 · answered Aug 11 '17 at 23:40

I have found a workaround to this problem. There might be a better solution to this problem, which I would love to hear about.

The workaround is to convert the one-hot encoding to categorical values. Of course, most of the time, we take our categorical values and convert to one-hot encoding. So just don't do this step.

Then, set the loss function to 'MultiClass'. This is the only loss function that catboost (and I think most gradient boosting packages) will support for multiclassification.

score 0 · Answer 2 · answered Nov 05 '20 at 09:13

0

catboost does the factors encoding automatically internally, no need to do it manually

answered Nov 05 '20 at 09:13

todorp

1
1

Does catboost support one-hot encoding?

2 Answers2