{
  "id": 587566,
  "title": "Single model or ensemble models",
  "url": "/competitions/cmi-detect-behavior-with-sensor-data/discussion/587566",
  "author_name": "",
  "post_date": "2025-07-01T15:59:21.345995800Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>In order to get consistent scores between CV and LB, I'm wondering if we should submit a single fold's model to the server or ensemble all the models together or it does not matter that much? I have a 0.06 gap between my CV and LB, and I am figuring out the reason</p>",
  "messages": [
    {
      "id": "3238163",
      "postDate": "07/01/2025 15:59:21",
      "content": "<p>In order to get consistent scores between CV and LB, I'm wondering if we should submit a single fold's model to the server or ensemble all the models together or it does not matter that much? I have a 0.06 gap between my CV and LB, and I am figuring out the reason</p>",
      "rawMarkdown": "In order to get consistent scores between CV and LB, I'm wondering if we should submit a single fold's model to the server or ensemble all the models together or it does not matter that much? I have a 0.06 gap between my CV and LB, and I am figuring out the reason",
      "votes": null
    },
    {
      "id": "3239007",
      "postDate": "07/02/2025 11:43:52",
      "content": "<p>Usually, an ensemble on all 5 folds will provide better correlation</p>",
      "rawMarkdown": "Usually, an ensemble on all 5 folds will provide better correlation",
      "votes": null
    },
    {
      "id": "3239008",
      "postDate": "07/02/2025 11:44:38",
      "content": "<p>If a post doesn’t interest you, you’re always free to scroll past. No need to be mean to others who want to learn.</p>",
      "rawMarkdown": "If a post doesn’t interest you, you’re always free to scroll past. No need to be mean to others who want to learn.",
      "votes": null
    },
    {
      "id": "3239044",
      "postDate": "07/02/2025 12:39:32",
      "content": "<p>Yup I think it will be more stable for the prediction.</p>",
      "rawMarkdown": "Yup I think it will be more stable for the prediction.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3239007,
      "author_name": "s617603",
      "author_url": "",
      "post_date": "07/02/2025 11:43:52",
      "content": "<p>Usually, an ensemble on all 5 folds will provide better correlation</p>",
      "votes": null,
      "replies": [
        {
          "id": 3239044,
          "author_name": "georgewang73",
          "author_url": "",
          "post_date": "07/02/2025 12:39:32",
          "content": "<p>Yup I think it will be more stable for the prediction.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 3239008,
      "author_name": "s617603",
      "author_url": "",
      "post_date": "07/02/2025 11:44:38",
      "content": "<p>If a post doesn’t interest you, you’re always free to scroll past. No need to be mean to others who want to learn.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3238163": "In order to get consistent scores between CV and LB, I'm wondering if we should submit a single fold's model to the server or ensemble all the models together or it does not matter that much? I have a 0.06 gap between my CV and LB, and I am figuring out the reason",
    "3239007": "Usually, an ensemble on all 5 folds will provide better correlation",
    "3239008": "If a post doesn’t interest you, you’re always free to scroll past. No need to be mean to others who want to learn.",
    "3239044": "Yup I think it will be more stable for the prediction."
  },
  "source": "meta"
}