ML Challenge Hack4Hack 2026

About

ML Challenge Hack4Hack 2026 is an online **machine-learning competition** in two rounds. The problem statement and dataset stay under wraps until **29 Oct**. **Round 1** is a one-day ranking round open to every registered team; the **top 100** go through to **Round 2**, three days of live model building on a fresh test set. **₹1 lakh** in prizes.

## How it works

1. **Register** as a team of 2 to 4 members, by **20 Oct, 11:59 PM IST**. Registration and Round 1 are free.

2. **Problem statement and dataset go live on 29 Oct, 10:00 AM IST** in your team workspace under Resources: the task, training data, test file (without labels), a sample submission and a data dictionary.

3. **Round 1, the one-day ranking round: 31 Oct, 10:00 AM to 11:59 PM IST.** Upload predictions from the **Leaderboard** tab of your workspace, up to **5 times**. Uploads are scored and the leaderboard, with your scores, updates every 4 hours.

4. **Top 100 announced on 1 Nov, 12:00 PM IST**, based on the final Round 1 ranking. Shortlisted teams are emailed and pay a **₹199 entry fee per team** from their workspace by **2 Nov, 11:59 PM IST** to confirm their place in Round 2.

5. **Round 2, live model development: 3 Nov, 10:00 AM to 5 Nov, 11:59 PM IST.** A new hidden test set and a fresh leaderboard. Up to **5 uploads per team per day**; the limit resets at midnight IST, and files that fail validation do not count.

6. **Submit your code** (GitHub repository or notebook link) from the Submission tab before Round 2 closes.

7. **Final ranking on 6 Nov** on the private part of the Round 2 test set, then code verification for the top teams. **Results on 8 Nov.**

## Submission format

Upload a CSV with two columns and exactly one row for every test id:

- **id**: the test sample id

- **label**: your prediction

## Scoring

- Uploads are scored with **macro F1**: the F1 score of each class, averaged equally, so rare classes matter as much as common ones.

- Each round's **leaderboard updates every 4 hours** and is scored on a public part of its test set.

- Each round's **final ranking** uses the hidden private part, based on each team's best public upload. Models that generalise win; overfitting the leaderboard does not.

- The Round 1 final ranking decides the top 100. Prizes are decided by the Round 2 final ranking.

## Prizes

- **1st place:** ₹50,000

- **2nd place:** ₹30,000

- **3rd place:** ₹20,000

- Certificates for every team that uploads a scored submission

## Who can participate

- College students from any year or branch, and working professionals

- Teams of 2 to 4 members

- Free to register and play Round 1; ₹199 per team to enter Round 2 after being shortlisted

Prizes

• 1st place — ₹50K
• 2nd place — ₹30K
• 3rd place — ₹20K

Perks

• Certificate for every team that uploads a scored submission
• Certificate of achievement for the top 3
• Free to register and play Round 1

Timeline

1. Registration and team formation (6 Oct 2026 – 20 Oct 2026): Register as a team of 2 to 4 members. Invite teammates by link or email until registration closes. Free.
2. Problem statement and dataset release (29 Oct 2026 – 30 Oct 2026): The problem statement, training data, test file (without labels), a sample submission and the data dictionary are published in your team workspace under Resources.
3. Round 1: one-day ranking round (31 Oct 2026 – 31 Oct 2026): Every registered team uploads predictions from the Leaderboard tab, up to 5 times. Uploads are scored with macro F1 and the leaderboard updates every 4 hours. The final Round 1 ranking decides the top 100.
4. Top 100 announced · Round 2 entry (1 Nov 2026 – 2 Nov 2026): The top 100 teams are emailed. Pay the ₹199 entry fee per team from your workspace by 2 Nov, 11:59 PM IST to confirm your place in Round 2.
5. Round 2: 3-day live model development (3 Nov 2026 – 5 Nov 2026): A new hidden test set and a fresh leaderboard for the top 100. Up to 5 uploads per team per day. Submit your code from the Submission tab before the round closes.
6. Final ranking and code verification (6 Nov 2026 – 7 Nov 2026): The Round 2 leaderboard switches to the private part of the test set. The top teams’ code is checked to make sure it reproduces their best upload.
7. Results (8 Nov 2026 – 8 Nov 2026): Winners announced on the hackathon page and by email.

Rules

1. Take part in a team of 2 to 4 members. Each person can be in only one team, with one account.
2. Train only on the dataset provided. Open-source libraries and publicly available pretrained weights are allowed; outside labelled data for the same task and hand-labelling the test set are not.
3. Upload predictions as a CSV with the columns id and label, with exactly one row for every test id.
4. Round 1 (31 Oct) is open to every registered team, with up to 5 uploads. Its final ranking decides the top 100 teams for Round 2.
5. Shortlisted teams confirm their Round 2 place by paying the ₹199 entry fee per team from their workspace by 2 Nov, 11:59 PM IST.
6. Round 2 (3–5 Nov) uses a new hidden test set. Each team gets 5 uploads per day; the limit resets at midnight IST, and files that fail validation do not count.
7. Each round is scored with macro F1. Scores and the leaderboard update every 4 hours, not instantly. Live leaderboards use a public part of the test set; final rankings use the hidden private part, based on each team’s best public upload. Prizes go by the Round 2 final ranking.
8. Before Round 2 closes, submit your code (a GitHub repository or notebook link) from the Submission tab. Prize-winning teams must be able to reproduce their best upload from that code.
9. Sharing code, predictions or labels between teams is not allowed and leads to disqualification.
10. The organizers’ decisions on shortlisting, verification and results are final.

Frequently asked questions

Is there a fee?

Registering and Round 1 are free. Teams shortlisted for Round 2 pay ₹199 per team (not per person) to confirm their place.

Can I take part alone?

No. Teams have 2 to 4 members. After registering, invite teammates by link or email until registration closes on 20 Oct.

When is the problem statement revealed?

On 29 Oct at 10:00 AM IST, together with the dataset, in your team workspace under Resources.

How are the top 100 chosen?

By the final Round 1 ranking: each team’s best upload on 31 Oct, scored on the private part of the Round 1 test set. Shortlisted teams are emailed on 1 Nov.

How do we pay the Round 2 fee?

From the Leaderboard tab of your team workspace, by UPI, card or netbanking, by 2 Nov, 11:59 PM IST. Any team member can pay.

What exactly do we upload?

A CSV of your predictions for the test set, with the columns id and label (one row per test id). A sample submission comes with the dataset.

How is it scored?

With macro F1: the F1 score of every class, averaged equally, so rare classes count as much as common ones.

How many times can we upload?

Up to 5 times in Round 1, and 5 times per team per day in Round 2. The limit resets at midnight IST, and uploads that fail validation do not use it up.

When do we see our score?

Scores are not shown instantly. The leaderboard and your own scores update every 4 hours from the start of each round (for Round 1: 2 PM, 6 PM and 10 PM IST, then the final ranking after the round closes). Files with format errors are rejected right away and do not use up an upload.

Why might my final rank differ from the live leaderboard?

Live leaderboards are scored on a public part of the test set. Final rankings use the private part, which rewards models that generalise instead of overfitting the leaderboard.

Do I need a GPU?

No. Free tiers such as Google Colab or Kaggle notebooks are enough.