GIZ NLP Agricultural Keyword Spotter
$7,000 USD
Classify audio utterances in Luganda and English from Uganda
476 data scientists enrolled, 139 on the leaderboard
11 September—29 November
Ends in 30 days
Multi Label or Multi class ?
published 10 Oct 2020, 16:07


First, Thanks you for releasing an audio competition !

I was looking into the data, and It seems to be a multi class problem, but when I look the example in evaluation, it seems the sum of the probability are above 1.0. Does that mean this competition is finally a multi label task ? Can we have multiple label for one audio sample ?

Hi :)

Each audio sample has ONE keyword as the label. Submission example shows format: floats for probabilities. Note there are also only three classes in that example. Sorry if it's misleading - this is still just a multi-class problem.

Should the probabilities add up to 1 or more than 1?

If they add up to more than one they'll be scaled during scoring if I remember correctly. You definitely won't get an error if they don't sum exactly to one.