0

I have to predict two labels and I have created a model using :

model = vw("--multilabel_oaa 50 --ngram E2 -b 24 -l2")

This returns two labels in 65% of predictions. But in the remaining cases it returns only one label or even three labels in some cases and also empty predictions.

[12, 44],
 [12, 47],
 [32, 44],
 [25, 44],
 [27, 28, 44],
 [44],
 [25, 44],
 [11, 44],
 []

I have tried with two separate models using oaa to predict the labels. But wanted to compare it with this approach but I am unable to get the expected predictions.

adithya krishnan
  • 33
  • 1
  • 1
  • 5
  • This is expected for multilabel. If you want a probability score for each label, use `--oaa N --loss_function=logistic --probabilities -p probs.txt`. – Martin Popel Sep 29 '20 at 07:43
  • Thanks for your response. The two labels are different categories in my case( Eg : Label 1 - the category and Label 2 - time bucket).I have encoded it to 1 to k for multilabel. Can you help on why are predictions not consistent on no of labels by using multilabel_oaa – adithya krishnan Sep 29 '20 at 10:37

0 Answers0