Lacuna Masakhane Parts of Speech Classification Challenge
Helping Africa
$7 000 USD
Completed (over 2 years ago)
Classification
Natural Language Processing
472 joined
101 active
Start
Jun 08, 23
Close
Sep 17, 23
Reveal
Sep 17, 23
First Zindi Comp - Confused about something:
Help · 8 Aug 2023, 22:08 · 2
Hello everyone,
Coming from kaggle, It's my very first Zindi comp I intend to participate in.
There is one thing I don't understand, wouldn't it be possible for someone who speaks Luo or Sestwana to label the data himself and thus to be able to make the perfect submission ?
There are 30k words in the test set, with a little group of individuals, this should be doable or at least partly and thus giving the user an unfair advantage as well as even having additionnal training data.
Please enlighten me if I'm wrong or missing something, looking foreward to participating in this comp.
J€ANMPIA
@JEANMPIA, welcome on board.
You're right to think in this direction but something important on Zindi for you to claim your top position is that your submitted training code put you back in that position.
This means that even if you find a way to cheat the system, your training code has to be submitted (NOT INFERENCE) and all you have done like training with prelabeled test data will be obvious.
Also, using the already trained model on the prelabeled test data is entirely not allowed as only freely available pre-trained weights are accepted. If you find a pre-trained model available for use that luckily has been trained with the test language? I think this is not discouraged as it was not trained on the provided test data and also not trained in-house without a working link of where you get it from, which will prove that it was indeed a freely available pre-trained weight.
Once again, welcome onboard, and happy coding!!!
Oh it makes sense now, thx for your time !