{
  "id": 519612,
  "title": "Any smart way to debug the submission?",
  "url": "/competitions/rsna-2024-lumbar-spine-degenerative-classification/discussion/519612",
  "author_name": "",
  "post_date": "2024-07-12T01:28:15.426170100Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>So, my submission runs OK in the small test set, and it runs 20 mins when submitting, but it seems it has an error somewhere. Is there any intelligent way to debug what is causing the error? Again, in the small output csv works like a charm</p>",
  "messages": [
    {
      "id": "2918041",
      "postDate": "07/12/2024 01:28:15",
      "content": "<p>So, my submission runs OK in the small test set, and it runs 20 mins when submitting, but it seems it has an error somewhere. Is there any intelligent way to debug what is causing the error? Again, in the small output csv works like a charm</p>",
      "rawMarkdown": "So, my submission runs OK in the small test set, and it runs 20 mins when submitting, but it seems it has an error somewhere. Is there any intelligent way to debug what is causing the error? Again, in the small output csv works like a charm",
      "votes": null
    },
    {
      "id": "2918053",
      "postDate": "07/12/2024 02:13:29",
      "content": "<p><a href=\"https://www.kaggle.com/lastmarchoftheents\" target=\"_blank\">@lastmarchoftheents</a>, one of the best way to debug is to run the entire pipeline on the train data and see if it gets to the submission.csv. <br>\nIt means you have a <code>Debug</code> flag in the beginning, so if it is true your dataset and model inference switch to <code>inference</code> on training data. It is a bit more hustle, though it is a proven way to approach the problem.</p>",
      "rawMarkdown": "lastmarchoftheents, one of the best way to debug is to run the entire pipeline on the train data and see if it gets to the submission.csv. \nIt means you have a `Debug` flag in the beginning, so if it is true your dataset and model inference switch to `inference` on training data. It is a bit more hustle, though it is a proven way to approach the problem.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2918053,
      "author_name": "sergiosaharovskiy",
      "author_url": "",
      "post_date": "07/12/2024 02:13:29",
      "content": "<p><a href=\"https://www.kaggle.com/lastmarchoftheents\" target=\"_blank\">@lastmarchoftheents</a>, one of the best way to debug is to run the entire pipeline on the train data and see if it gets to the submission.csv. <br>\nIt means you have a <code>Debug</code> flag in the beginning, so if it is true your dataset and model inference switch to <code>inference</code> on training data. It is a bit more hustle, though it is a proven way to approach the problem.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2918041": "So, my submission runs OK in the small test set, and it runs 20 mins when submitting, but it seems it has an error somewhere. Is there any intelligent way to debug what is causing the error? Again, in the small output csv works like a charm",
    "2918053": "lastmarchoftheents, one of the best way to debug is to run the entire pipeline on the train data and see if it gets to the submission.csv. \nIt means you have a `Debug` flag in the beginning, so if it is true your dataset and model inference switch to `inference` on training data. It is a bit more hustle, though it is a proven way to approach the problem."
  },
  "source": "meta"
}