Back to all discussions

  •   about 7 years ago

Data with multiple classes

When scoring the test data, do we get any kind of partial score for samples with multiple classes? E.g., 3850.png has bananas and tomatoes (bananas look more prominent to me), and my classifier has bananas while the tsv has tomatoes. I don't see the same photo in another file in the tsv that's bananas, so it doesn't look like it's classified twice.

  • 3 comments

  • Manager   •   about 7 years ago

    Hi Jayen!

    Thanks for your message. :) This is a natural dataset where the data was labeled by our CS agents during processing of customer feedback. This means it will contain some noise and we encourage to find creative ways to handle this.

    Good luck! ;)

    Best,

    Willem

  •   •   about 7 years ago

    I can understand that the training data has this noise. That is very realistic. What I wonder is if the scoring test set has noise? Because then we would get penalised for having the right classification.

  • Manager   •   about 7 years ago

    The noise in the test set is very minimal, but we cannot confirm it is completely noise-free. In the case that there are 2 teams with almost the same score we can do a double pass to check.

Comments are closed.