Use version 2 of the Q2Q dataset

[wael34218]

The new data can be found under this link:
https://s3-eu-west-1.amazonaws.com/nsurlworkshop/q2q_similarity_workshop_v2.tsv

This repository will be released for the public, test data shouldn't be in the repo nor its git history.

[hseelawi]

Done. Please check and let me know if all is good.

[hinnaweali]

Hello,

Could you please share with us the public and private test sets so we can have a fair comparison results with the group that won the competition in 2019?

[hseelawi]

Hello @hinnaweali,

I am attaching a zip file that contains all the files we produced for this workshop. It has both the public (train.tsv) and private (test.tsv) datasets.

nsurl-2019-task8.zip

[hinnaweali]

Thank you @hseelawi.

I downloaded those files from Kaggle website. You mentioned there, the private leaderboard was calculated with approximately 70% of the test data while the public leaderboard was calculated with approximately 30% of the test data.
I need exactly those splits to evaluate my model on both 30% and 70% test data sets.
Best,

[hseelawi]

Hello @hinnaweali again,

Actually we can't get the exact split as this is handled by kaggle itself, and is not available even for us the organizers to download. However, what matters is the private score, which you can obtain by submitting a late submission to the leaderboard directly, at the following link: https://www.kaggle.com/c/nsurl-2019-task8/submit.

[Update]

You can also find the public score after you do a late submission as well.

[hinnaweali]

Hi @hseelawi,

This is really a nice tip.
Thank you

[hseelawi]

You are most welcome. Please let us know should you need any further information or help.