{
  "fetched_at": "2026-10-19T10:45:05Z",
  "source_url": "https://www.kaggle.com/competitions/codata2019challenge/data",
  "pages": [
    {
      "name": "rules",
      "content": "There are no rules to this competition. Using any method you wish, your goal is to take a set of Time Traces (time series) from a fixed **but unknown* experimental condition (0, 1, or 2), and predict that hidden experimental condition. \n\nCorrect predictions are worth one point. Incorrect predictions are worth zero points. ",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "data-description",
      "content": "# Data Description\n\nEach row of the data file `training.csv` has columns:\n\n - `well_id` :  the ID of the well, an integer.\n - `treatment_condition` : the treatement condition, an integer. Note that in any given well, **all of the treatment conditions will be the same, since those cells are physically located together in the same conditions**.\n    - 0 = DMSO \n    - 1 = CCCP\n    - 2 = Apramycin \n - `ts_1`, `ts_2`, ... , `ts_600` : 600 timeseries steps, representing the timeseries collected from that cell. \n\nThe file `test.csv` is identical in format to the file `training.csv` except that the column for `treatment_condition` is empty. It is up to you to make this classification!\n\nAgain, all traces from the same well will be in the same treatment class. THIS WILL ALSO BE TRUE OF THE CLASSIFICATION TASKS IN TEST DATA. ",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "# Abstract\n\nToday's microscopes, combined with video recording and automation, are capable of real-time observations of microscopic phenomena. In this challenge, we'll use timeseries data, recorded from real cells through exactly such a microscope, to classify which drug the cells were exposed to. \n\n# Data Description\n\nEach row of the data file `training.csv` has columns:\n\n - `well_id` : the ID of the well, an integer.\n - `treatment_condition` : the treatement condition, an integer. Note that in any given well, **all of the treatment conditions will be the same, since those cells are physically located together in the same conditions**.\n\t- 0 = DMSO \n\t- 1 = CCCP\n\t- 2 = Apramycin \n - `ts_1`, `ts_2`, ... , `ts_600` : 600 timeseries steps, representing the timeseries collected from that cell. \n\nThe file `test.csv` is identical in format to the file `training.csv` except that the column for `treatment_condition` is empty. It is up to you to make this classification!\n\nAgain, all traces from the same well will be in the same treatment class. THIS WILL ALSO BE TRUE OF THE CLASSIFICATION TASKS IN TEST DATA. \n\n# Data Sharing\n\nData are not to be shared outside of the DS Team + collaborators. Data and notes may not be posted online in any public-facing repo under any circumstances. This is a private competition. ;)\n\n# Goal and evaluation criteria\n\nSuppose that you are provided with a set of time traces from a well. The well has either DMSO, CCCP, or Apramycin in it, corresponding to a label of 0, 1, or 2.  \n\nImagine that you do not know this label or the compound. All that you have is a set of those time traces from the unknown well. \n\n**Can you make an accurate classification of 0, 1, or 2, given a set of time-series from an unlabeled well?**\n\nYour goal is to use the training data provided by Prof. Kralj to make a classification about the well identity. Later in the semester, we will provide you with *exactly* this sort of unlabeled data, and ask you to make exactly this kind of classification.\n\nEvaluation scores will be simply: fraction of timeseries predicted correctly. \n\nYou may use any off-the-shelf packages for analyzing time series if you wish, or you can create any kind of custom features that you like. Everything is on the table. The only requirement is that you must be able to explain and/or cite the ideas, methods, or packages that you use. \n\n# Acknowledgements\n\nData courtesy of Prof. Joel Kralj,\nUniversity of Colorado Boulder,\nDepartment of Molecular, Cellular, and Developmental Biology, \nand BioFrontiers Institute.\n\nDirect all questions to Prof. Dan Larremore,\nUniversity of Colorado Boulder,\nDepartment of Computer Science,\nand BioFrontiers Institute.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "The evaluation metric for this competition is **Classification Accuracy**. Each well has a particular treatment condition, 0, 1, or 2. Accuracy is simply the fraction of wells that you have correctly predicted., \n\n# Submission Format\nThe submission CSV should contain two columns with headers: \n\n - `Id` : the Id of the well being predicted. For test data, these IDs range from `451` to `576`. \n - `Category` : your prediction of the treatment condition: 0, 1, or 2. \n\nAn example submission file, in which I have placed uniform random guesses, can be found in `submission_example.csv` on the Data page. ",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}