mohanty

[Final Week!] Leaderboard process, Winner announcement timeline, duplicate accounts

The real Lulz comes in when everyone realises that the @alfarzan account has been powered by a GPT-3 powered bot all along !
Not giving any hints

View post 4

What's the etiquette on sharing code?

Hi @simon_coulombe,

We highly encourage participants to share their code.
Please feel free to create a separate repository or a google colab notebook and share the link here on the forums.

I believe @shivam and team had also been working on a better way to share code more natively on the platform.

Best,
Mohanty

View post 3

Tips on creating a successful submission

@jyotish: Can we basically have a custom base image where all the common dependencies are already installed ?
It will hopefully both reduce the image build times and also help us avoid back and forth many such dependency related issues.

View post 2

Video with "en passant" capture

@demarsylvain: Looking at this, the notation seems to be to add a .p..
I will let @Shubhamaicrowd confirm if we have instances of the same in our dataset.

View post 1

Video with "en passant" capture

@demarsylvain: You can always generate one

View post 1

About the datasets for WinPrediction

@victorkras2008: Thanks for pointing these out. We are on it. And indeed, the some of the labels should also be wrong in the test set. We are working on correcting the affected data points, and then re-evaluating all the submissions.
@ayushivani is on it !

View post 0

About white passed pawn

Thanks @Shubhamaicrowd !
It’s a bit weird though, I would expect python-chess to have such a silly bug. Nor do I understand why they wouldn’t consider the side on pawn promotion ?

In any case, if anyone has an understanding of why this is hapenning, please do let us know.
Else I suggest to include a rule based correction for this case and update the dataset for consistency.

Best,
Mohanty

View post 0

About white passed pawn

@victorkras2008: Hi, we are acknowledging this issue and are investigating this closely as we speak, we will post updates on this soon. Thanks for notifying us of the bug.

Best,
Mohanty

View post 0

Clarification about final model selection

Hi @rejulien,

For the final leaderboard, the best model on the whole test set will be used.

Best,
Mohanty

View post 0

Problem with some SMILES

Hey,

We are on this.
We will get back to you soon with updates from our end.
@shivam is coordinating the efforts from our end.

Best,
Mohanty

View post 0

Supplementary Training Data Release for Round-3

Dear all,

We have released some Supplementary Training data for Round-3.
This data contains smell words for 388 molecules.
Please note that, for consistency, the smell words belong to the global vocabulary used in Round-1 and Round-2 of the competition and hence the smell sentences contain smell words that may not be in the Round-3 vocabulary.

The data can be accessed in the Resources Section of the challenge.

Best of Luck,
Mohanty

View post 0

Round3-vocabulary.txt contains strange vocabularies

@hjuinj: Yes, we have.

View post 1

Submission Error in Non-debug Mode

@hjuinj: the test set for this round contains numerous molecules which are not present in the test sets of previous round. Hence precomputed features will not work here.
You will have to ensure you can do feature computation on the fly before you make the predictions

View post 0

Round3-vocabulary.txt contains strange vocabularies

@hjuinj: The train.csv is something that we have consistently used across the previous rounds, and hence changing the same can create some confusion.
If filtering the invalid vocab items for round 3 is adding friction, we would be very happy to upload a filtered version.
We will update this thread as soon as its done.

Best,
Mohanty

View post 0

Round3-vocabulary.txt contains strange vocabularies

@koupenchan: Thanks for pointing this out. We have updated the round3-vocabulary.txt file, and also re-evaluated all the previous submissions.

Best of luck,
Mohanty

View post 1

Build failure for custom conda environment. Frustrating and time consuming to debug the build failures

Hi @mario_lovric,

I am sorry about the hurdles in packaging your software environment. Have you managed to write a Dockerfile which can do the predictions at your end ?
Someone from our team would anyway reach out to assist you in the process.

Best,
Mohanty

View post 0

Build failure for custom conda environment. Frustrating and time consuming to debug the build failures

@justmahesh: The need for code based submissions is to allow us to evaluate your models on test sets that cannot be released publicly because of legal constraints. And the need for submitting code does add a barrier to entry which we acknowledge, and we are willing to assist all the participants in packaging their code for making a valid submissions.

We will follow up on this particular issue, but we also have office hours (@jyotish : can you please share details) to help all the participants on a one on one basis.

Best,
Mohanty

View post 0

#1 Solution to TIMSER Blitz5

@boussif_oussama: “AIcrowders” has a weird ring to it. At AIcrowd, we prefer “AIcrew” to refer to the proud members of the community

View post 3

Food Recognision Challenge | Round 4 Announcement 🍱

@markpotanin: We will release some more supplementary training data shortly. Thanks for your patience.

View post 0

Questions about leaderboard / private set

@ktverdov: This round of the food benchmark does not have a separate validation set, and hence the scores on the leaderboard are the final scores for this round. We release new test sets at the end of each round. In the subsequent rounds we will ensure to have separate public and private scores.

The recall score will be used as a tie breaker. So if two submissions have the exact same (upto 3 decimal points) precision then their order will be determined by the recall scores.

Best,
Mohanty

View post 2