{
  "id": 335920,
  "title": "Is training supposed to be run online?",
  "url": "/competitions/hubmap-organ-segmentation/discussion/335920",
  "author_name": "",
  "post_date": "2022-07-08T13:25:14.719799Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I apology if this topic is old.  I haven't participated in Kaggle competitions for several years and I found myself in total loss.  Could someone please point to me which of the following is the intention of the organizers?</p>\n<p>A. The public training data (including one testing image)  is the final training data.  We are supposed to train and prepare the model offline.  We then upload the model and a thin predicting notebook to run predictions online.</p>\n<p>B. The submitted notebook is supposed to contain both training and testing code.  We cannot assume that when the notebook is run it sees the identical training data as the public data.  (We should therefore still upload the model and finetune with the online data.)</p>",
  "messages": [
    {
      "id": "1848249",
      "postDate": "07/08/2022 13:25:14",
      "content": "<p>I apology if this topic is old.  I haven't participated in Kaggle competitions for several years and I found myself in total loss.  Could someone please point to me which of the following is the intention of the organizers?</p>\n<p>A. The public training data (including one testing image)  is the final training data.  We are supposed to train and prepare the model offline.  We then upload the model and a thin predicting notebook to run predictions online.</p>\n<p>B. The submitted notebook is supposed to contain both training and testing code.  We cannot assume that when the notebook is run it sees the identical training data as the public data.  (We should therefore still upload the model and finetune with the online data.)</p>",
      "rawMarkdown": "I apology if this topic is old.  I haven't participated in Kaggle competitions for several years and I found myself in total loss.  Could someone please point to me which of the following is the intention of the organizers?\n\nA. The public training data (including one testing image)  is the final training data.  We are supposed to train and prepare the model offline.  We then upload the model and a thin predicting notebook to run predictions online.\n\nB. The submitted notebook is supposed to contain both training and testing code.  We cannot assume that when the notebook is run it sees the identical training data as the public data.  (We should therefore still upload the model and finetune with the online data.)",
      "votes": null
    },
    {
      "id": "1848545",
      "postDate": "07/08/2022 17:59:17",
      "content": "<p>You can only upload the inference and train your model offline, the purpose of a code submission is to hide the test data</p>",
      "rawMarkdown": "You can only upload the inference and train your model offline, the purpose of a code submission is to hide the test data",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1848545,
      "author_name": "enric1296",
      "author_url": "",
      "post_date": "07/08/2022 17:59:17",
      "content": "<p>You can only upload the inference and train your model offline, the purpose of a code submission is to hide the test data</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1848249": "I apology if this topic is old.  I haven't participated in Kaggle competitions for several years and I found myself in total loss.  Could someone please point to me which of the following is the intention of the organizers?\n\nA. The public training data (including one testing image)  is the final training data.  We are supposed to train and prepare the model offline.  We then upload the model and a thin predicting notebook to run predictions online.\n\nB. The submitted notebook is supposed to contain both training and testing code.  We cannot assume that when the notebook is run it sees the identical training data as the public data.  (We should therefore still upload the model and finetune with the online data.)",
    "1848545": "You can only upload the inference and train your model offline, the purpose of a code submission is to hide the test data"
  },
  "source": "meta"
}