{
  "fetched_at": "2026-10-20T07:52:42Z",
  "source_url": "https://www.kaggle.com/competitions/dlsprint3/data",
  "pages": [
    {
      "name": "data-description",
      "content": "# Public Train and Test Set \n\nThe training set consists of 209 problems from Math Olympiad and their answers. All problems have non-negative integer answers. All problems are text-only with mathematical notation in LaTeX. Although some problems may involve geometry, diagrams are not used in any problem.\n\nBecause of the limited number of problems available, we are taking special precautions to secure the test set against probing, hardcoding, overfitting. Among other things, during the Phase 1 of the competition, the test set will comprise only the 100  problems. \n\nIn Phase II, we will rerun submissions on a holdout set which is **completely different** from the training and test dataset. You should attempt to make sure your submission will complete successfully on the holdout private set problems. This may mean ensuring your submission is runnable in Kaggle environment and runs within permissible space and time limit. \n\n## Files\n\n*   **train.csv** - the training set\n*   **test.csv** - the test set\n*   **sample_submission.csv** - a sample submission file in the correct format\n\n## Columns\n### Train Set\n\n*   `Id` - Unique id for the problem\n*   `Problem` - Problem statement\n*   `Answer` - Integer answer of the problem\n\n### Test Set\n\n*   `Id` - Unique id for the problem\n*   `Problem` - Problem statement\n\n### Submission Format\n*   `Id` - Unique id for the problem\n*   `Answer` - Predicted answer (**must be an Integer**)\n\n\n# Holdout Set \nThe holdout set will be used Phase II. The set consists of 100 new problems created in collaboration with BDMO. All the problems have non-negative integer answers as well. They will be text-only with mathematical notation in LaTeX and they will not contain diagrams. The difficulty of the problems are moderately higher than the problems in the public train and test set.\n\n## Files\n*   **test.csv** - the test set\n\n## Columns\n*   `Id` - Unique id for the problem\n*   `Problem` - Problem statement\n\n### Submission Format\n*   `Id` - Unique id for the problem\n*   `Answer` - Predicted answer (**must be an Integer**)",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "The ability to understand and solve math problems is a foundational skill for AI, paving the way for innovations across fields such as science, engineering, and financial modeling. Current AI models are still limited in their ability to tackle mathematical challenges in a low-resource language like Bengali. By addressing problems from the Bengali Math Olympiad, this competition challenges participants to overcome these limitations and expand AI’s reach.\n\nThis competition is the first of its kind in the Bengali language. By participating, you'll contribute to advancing AI capabilities in mathematical reasoning, bringing us closer to developing models that can tackle high-level problems with precision and accuracy. Your models which can reason accurately in Bengali can help and guide students and researchers alike to learn and solve more complex problems. \n\nThe event will take place in two phases. \n## Phase I\nTeams will compete on Kaggle public leaderboard using the test set. At the end of phase I, teams will need to submit their inference notebooks.\n\n## Phase II\nThe inference notebook will be rerun on a hidden test set. The hidden test set will contain **100** new unseen problems created in collaboration with BDMO. The final evaluation will be done based on both Phase I and Phase II score. \n\nParticipants are highly encouraged to develop their pipeline in Bengali. Such approaches to the problem will be rewarded with novelty score.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Submissions are evaluated based on exact match between correct answers and predicted answers. No partial credit will be given. Each problem has an integer answer.\n\n\n## Submission File\nFor each ID in the test set, you must predict a answer to the math problem, which must be a non-negative integer. The file should contain a header and have the following format:\n\n    ID,Answer\n    2,0\n    5,0\n    6,0\n    ...",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "# Models and Training:\n- Participants may use pre-trained models as long as they are open-source and appropriately licensed. \n- Fine-tuning of models is permitted.\n- **API usage of any kind is strictly prohibited.** (i.e. OpenAI or Claude).\n- Using external infrastructure for model training is allowed.\n\n# Translation\n- Although a Bengali pipeline is highly encouraged, translation  is allowed.\n- Translation must be done using a model to which the aforementioned rules also apply.\n- Translation model and training code must be submitted.\n- Translation on the test set must be done in the inference notebook.\n\n# Use of External Data:\n\n- Participants are allowed to use external data only if one of the following following is applicable-\n-- The data is publicly available.\n-- The data is curated from publicly available sources.\n-- The data is created by the participants during the competition runtime.\n- Any external data used must be declared and made public at the time of submission, with appropriate citations.\n\n# Submission:\n- Participants must submit the code for their training and inference pipeline.\n- Participants must submit their entire training methodology and ensure reproducibility. \n- The submitted inference notebook must be runnable on kaggle (CPU or GPU) within kaggle runtime [limits](https://www.kaggle.com/docs/notebooks) and reproduce the same results.\n- Submissions will be rerun on the public test set and on the hidden test set after the Phase I concludes.\n- Participants must ensure their final submission is well-documented and can be replicated for scoring.\n- Participants must make their final submission notebooks public.\n- Participants need to submit a paper in IEEE format describing their solution.\n\n# Submission Limits:\n\n- Each participant or team can submit up to 5 entries per day during the competition.\n- Submissions are evaluated on the leaderboard using 100 problems in the public test set.\n- After the competition, the final score will be determined by evaluating submissions on both public test set and the hidden test set.\n\n# Evaluation Metric:\n\n- Submissions will be evaluated based on accuracy. The answer to each problem must be predicted to be an integer.\n- Partial credit will not be given; the prediction must exactly match the true answer.\n\n# Eligibility:\n\n- The competition is open to individuals and teams. However, participants must comply with all legal requirements and competition rules.\n- Participants must register for the competition with a valid account on the Kaggle.\n\n# Originality and Plagiarism:\n\n- All submitted code and models must be the original work of the participant or team.\n- Participants may not use, copy or share any part of another participant’s code or solution, unless made public before the end of the competition.\n\n\n# Team Formation:\n- Teams may consist of up to 5 members. \n- At most 1 member of each team can be a professional or expert.\n- The remaining members must be undergraduate students.\n- Merging of teams is allowed until the team merge deadline of the competition. After this point, no new team merges will be accepted.\n\n# Code of Conduct:\n- Participants are expected to uphold the spirit of the competition by following the rules and ensuring fairness and transparency.\n- Any attempts to manipulate the leaderboard or exploit system vulnerabilities will result in disqualification.\n- Violations of these rules may result in disqualification from the competition.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": "Artificial Intelligence is no longer a concept from science fiction. Modern AI has a huge potential to change the world. And it all starts with intelligence. What better way to make AI models intelligent than to make them good at problem solving? \nIn this competition you will be creating one of the first AI models that can solve Bengali Math Problems. Your work will establish benchmarks in complex mathematical reasoning in the Bengali language.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Timeline",
      "content": "**Phase 1 Date & Place:** October 7, 2024 to October 31, 2024. Hosted online on the Kaggle platform. \n**Phase 2 Date & Place:** November 6 ( tentative ) , 2024. Hosted in the Department of CSE, BUET premises.\n\n**Competition and Registration Start:** October 4, 2024\n**Registration Deadline:** October 24, 2024\n**Team Formation/Merge Deadline:** October 24, 2024\n**Phase 1 Submission Deadline:** October 31, 2024\n\n**Registration Form Link:** [Click Here](https://tinyurl.com/dlsprint-reg-form)\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Final Scoring",
      "content": "Teams will be evaluated in the following criteria.\n| Criteria | Weight |\n| --- | --- |\n| Phase I: Leaderboard | 20% |\n| Phase II: Hidden Set Score | 40% |\n| Novelty| 30% |\n| Presentation and paper | 10% |\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}