{
  "fetched_at": "2026-10-24T15:59:52Z",
  "source_url": "https://www.kaggle.com/competitions/dla-4-a-test/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The data provided for this challenge is a subset of the same one used by Kratzert et al. to show that an LSTM approach significantly outperforms the NOAA model.\n\n**File descriptions**\n- **DLA_Project-A_challenge_train.csv** - 20 years of data for 100 of the CAMELS U.S. basins; use this dataset for both training and validation, partitioning it however you think is best (e.g. to avoid overfitting).\n- **DLA_Project-A_challenge_test.csv** - 10 years of data for the same 100 CAMELS U.S. basins provided in the training and validation set.\n- **DLA_Project-A_challenge_test_unseen_basins.csv** - Data for 10 U.S. basins not included in the training and validation set; to be used for more fully evaluating model generalization performance.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "This challenge is a continuation and extension of the Project-A Exercises notebook.\n\nDo the best job you can to predict discharge across 100+ U.S. basins extracted from the Catchment Attributes and Meteorology for Large-sample Studies (CAMELS) dataset.\n\nAdditional information and a sample template are provided in the Project-A Challenge notebook we are distributing from our CodeCommit repo. Information on how to get set up for the challenge is provided in the notebook and also in our \"Project A\" Teams channel.\n\nIMPORTANT: As questions or issues arise during the challenge, we will use Teams (not Kaggle) to communicate and provide updates.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Model performance will be an important part of the overall evaluation for the Project-A Challenge, but not the only factor. Several other factors that address Knowledge Sharing will also be considered. (Point scheme will be announced at a later date via Teams) \n<br>\n- Leaderboard results at the time the Kaggle competition closes, as measured by R-squared; plus assessment of results for the 10 additional (unseen) basins. \n<br>\n- Quality of code and documentation provided in the notebook, assessed from the perspective of assisting others who may reuse your notebook in the future. \n<br>\n- Ability to effectively communicate results and insights to a broad, non-expert audience.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "- While we strongly recommend that you use LSTM layers for the challenge, any network topology is acceptable.\n- You MUST use the training/validation and test datasets as instructed.\n- You MUST train a global model. You should not train a model specific to a particular basin or subset of basins.\n- Beyond that, you can do whatever you want… add layers, change the head layer, tune hyperparameters, change loss functions, use ensembles, etc.\n- Report any blocker issues in our \"Project A\" Teams Channel ASAP.\n- Collaborate with your teammate and have fun!\n- Good luck!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}