{
  "id": 314679,
  "title": "The upper bound for ensembling？",
  "url": "/competitions/happy-whale-and-dolphin/discussion/314679",
  "author_name": "",
  "post_date": "2022-03-24T00:16:25.436263700Z",
  "votes": 5,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Noticing the successive improvements of the ensemble notebooks, I wonder what's the maximum of this hack.</p>\n<p>I think there should be some theoretical upper bound.</p>\n<p>But at least we can see the experiment results here.</p>",
  "messages": [
    {
      "id": "1733020",
      "postDate": "03/24/2022 00:16:25",
      "content": "<p>Noticing the successive improvements of the ensemble notebooks, I wonder what's the maximum of this hack.</p>\n<p>I think there should be some theoretical upper bound.</p>\n<p>But at least we can see the experiment results here.</p>",
      "rawMarkdown": "Noticing the successive improvements of the ensemble notebooks, I wonder what's the maximum of this hack.\n\nI think there should be some theoretical upper bound.\n\nBut at least we can see the experiment results here.",
      "votes": null
    },
    {
      "id": "1733027",
      "postDate": "03/24/2022 00:36:38",
      "content": "<p>My score is ensembling of public kernels. It is possible to go 0.20 higher than 0.752 just using public kernels. It is mainly about the diversity of models but usually, we are bounded by the quality of the model which directly depends on the quality of the data. This is why changes in the dataset can greatly increase or decrease CV scores.</p>",
      "rawMarkdown": "My score is ensembling of public kernels. It is possible to go 0.20 higher than 0.752 just using public kernels. It is mainly about the diversity of models but usually, we are bounded by the quality of the model which directly depends on the quality of the data. This is why changes in the dataset can greatly increase or decrease CV scores.",
      "votes": null
    },
    {
      "id": "1733094",
      "postDate": "03/24/2022 02:16:18",
      "content": "<p>Thank you. Great explanation！</p>",
      "rawMarkdown": "Thank you. Great explanation！",
      "votes": null
    },
    {
      "id": "1739336",
      "postDate": "03/30/2022 01:11:05",
      "content": "<p>Good question.</p>",
      "rawMarkdown": "Good question.",
      "votes": null
    },
    {
      "id": "1739338",
      "postDate": "03/30/2022 01:13:45",
      "content": "<p>Thank you     </p>",
      "rawMarkdown": "Thank you",
      "votes": null
    },
    {
      "id": "1739351",
      "postDate": "03/30/2022 01:27:04",
      "content": "<p>We can try to iterate the weights</p>",
      "rawMarkdown": "We can try to iterate the weights",
      "votes": null
    },
    {
      "id": "1739352",
      "postDate": "03/30/2022 01:27:45",
      "content": "<p>Thx for the idea. </p>",
      "rawMarkdown": "Thx for the idea.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1733027,
      "author_name": "outwrest",
      "author_url": "",
      "post_date": "03/24/2022 00:36:38",
      "content": "<p>My score is ensembling of public kernels. It is possible to go 0.20 higher than 0.752 just using public kernels. It is mainly about the diversity of models but usually, we are bounded by the quality of the model which directly depends on the quality of the data. This is why changes in the dataset can greatly increase or decrease CV scores.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1733094,
          "author_name": "alanhabrony",
          "author_url": "",
          "post_date": "03/24/2022 02:16:18",
          "content": "<p>Thank you. Great explanation！</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1739336,
      "author_name": "upvoti",
      "author_url": "",
      "post_date": "03/30/2022 01:11:05",
      "content": "<p>Good question.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1739338,
          "author_name": "alanhabrony",
          "author_url": "",
          "post_date": "03/30/2022 01:13:45",
          "content": "<p>Thank you     </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1739351,
      "author_name": "upvoti",
      "author_url": "",
      "post_date": "03/30/2022 01:27:04",
      "content": "<p>We can try to iterate the weights</p>",
      "votes": null,
      "replies": [
        {
          "id": 1739352,
          "author_name": "alanhabrony",
          "author_url": "",
          "post_date": "03/30/2022 01:27:45",
          "content": "<p>Thx for the idea. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1733020": "Noticing the successive improvements of the ensemble notebooks, I wonder what's the maximum of this hack.\n\nI think there should be some theoretical upper bound.\n\nBut at least we can see the experiment results here.",
    "1733027": "My score is ensembling of public kernels. It is possible to go 0.20 higher than 0.752 just using public kernels. It is mainly about the diversity of models but usually, we are bounded by the quality of the model which directly depends on the quality of the data. This is why changes in the dataset can greatly increase or decrease CV scores.",
    "1733094": "Thank you. Great explanation！",
    "1739336": "Good question.",
    "1739338": "Thank you",
    "1739351": "We can try to iterate the weights",
    "1739352": "Thx for the idea."
  },
  "source": "meta"
}