{
  "id": 232396,
  "title": "Model-centric vs data-centric",
  "url": "/competitions/hpa-single-cell-image-classification/discussion/232396",
  "author_name": "narsil (jobs-in-data.com)",
  "post_date": "2021-04-13T15:01:21.234000",
  "votes": 22,
  "comment_count": 6,
  "views": 0,
  "content": "<p><a href=\"https://www.youtube.com/watch?v=qnCJvQL9wcg\" target=\"_blank\">https://www.youtube.com/watch?v=qnCJvQL9wcg</a></p>\n<p>I recently came across this fantastic talk by Andrew NG about the model-centric approach vs data-centric approach and I find it particularly relevant for this competition.</p>\n<p>My main lesson from this one is: do not try to fix with model architecture something that should be fixed by other elements of ML pipeline. </p>",
  "messages": [
    {
      "id": 1272546,
      "postDate": "2021-04-13T15:01:21.233Z",
      "content": "<p><a href=\"https://www.youtube.com/watch?v=qnCJvQL9wcg\" target=\"_blank\">https://www.youtube.com/watch?v=qnCJvQL9wcg</a></p>\n<p>I recently came across this fantastic talk by Andrew NG about the model-centric approach vs data-centric approach and I find it particularly relevant for this competition.</p>\n<p>My main lesson from this one is: do not try to fix with model architecture something that should be fixed by other elements of ML pipeline. </p>",
      "rawMarkdown": "https://www.youtube.com/watch?v=qnCJvQL9wcg\n\nI recently came across this fantastic talk by Andrew NG about the model-centric approach vs data-centric approach and I find it particularly relevant for this competition.\n\nMy main lesson from this one is: do not try to fix with model architecture something that should be fixed by other elements of ML pipeline. ",
      "votes": 22
    },
    {
      "id": 1659461,
      "postDate": "2022-01-21T20:24:02.933Z",
      "content": "<p>NeurIPS Data-Centric AI workshop in December 2021 also showed that the data-centric approach is becoming more and more popular; it's fantastic! </p>",
      "rawMarkdown": "NeurIPS Data-Centric AI workshop in December 2021 also showed that the data-centric approach is becoming more and more popular; it's fantastic! ",
      "votes": 1
    },
    {
      "id": 1273174,
      "postDate": "2021-04-14T06:26:58.190Z",
      "content": "<p>I think data-centric approach should be better</p>",
      "rawMarkdown": "I think data-centric approach should be better"
    },
    {
      "id": 1273068,
      "postDate": "2021-04-14T04:35:43.053Z",
      "content": "<p>What are the other elements of ML pipeline?</p>",
      "rawMarkdown": "What are the other elements of ML pipeline?",
      "replies": [
        {
          "id": 1273077,
          "postDate": "2021-04-14T04:44:05.123Z",
          "content": "<p>Probably that is the secret until the end of this competition!</p>",
          "rawMarkdown": "Probably that is the secret until the end of this competition!"
        },
        {
          "id": 1273215,
          "postDate": "2021-04-14T07:05:16.033Z",
          "content": "<p>I will gladly share everything after the end of the competition. No single source of value here - many elements coming together contribute to our score. I am honestly amazed by how fast you (and other players) were able to increase your score - in my case, it was much longer and needed many, many iterations until I got at least something right.</p>",
          "rawMarkdown": "I will gladly share everything after the end of the competition. No single source of value here - many elements coming together contribute to our score. I am honestly amazed by how fast you (and other players) were able to increase your score - in my case, it was much longer and needed many, many iterations until I got at least something right.",
          "votes": 5
        },
        {
          "id": 1273284,
          "postDate": "2021-04-14T08:06:11.937Z",
          "content": "<p>So far, I am still solo. When I see you get 0.56+, how regretfully I didn't join you. Hah, good luck.</p>",
          "rawMarkdown": "So far, I am still solo. When I see you get 0.56+, how regretfully I didn't join you. Hah, good luck.",
          "votes": 3
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 1659461,
      "author_name": "Evgeniya Sukhodolskaya",
      "author_url": "",
      "post_date": "2022-01-21T20:24:02.933000",
      "content": "<p>NeurIPS Data-Centric AI workshop in December 2021 also showed that the data-centric approach is becoming more and more popular; it's fantastic! </p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1273174,
      "author_name": "ZavodRobotov",
      "author_url": "",
      "post_date": "2021-04-14T06:26:58.190000",
      "content": "<p>I think data-centric approach should be better</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 1273068,
      "author_name": "Correlation",
      "author_url": "",
      "post_date": "2021-04-14T04:35:43.053000",
      "content": "<p>What are the other elements of ML pipeline?</p>",
      "votes": 0,
      "replies": [
        {
          "id": 1273077,
          "author_name": "Shai",
          "author_url": "",
          "post_date": "2021-04-14T04:44:05.123000",
          "content": "<p>Probably that is the secret until the end of this competition!</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1273215,
          "author_name": "narsil (jobs-in-data.com)",
          "author_url": "",
          "post_date": "2021-04-14T07:05:16.033000",
          "content": "<p>I will gladly share everything after the end of the competition. No single source of value here - many elements coming together contribute to our score. I am honestly amazed by how fast you (and other players) were able to increase your score - in my case, it was much longer and needed many, many iterations until I got at least something right.</p>",
          "votes": 5,
          "replies": []
        },
        {
          "id": 1273284,
          "author_name": "Correlation",
          "author_url": "",
          "post_date": "2021-04-14T08:06:11.937000",
          "content": "<p>So far, I am still solo. When I see you get 0.56+, how regretfully I didn't join you. Hah, good luck.</p>",
          "votes": 3,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1272546": "https://www.youtube.com/watch?v=qnCJvQL9wcg\n\nI recently came across this fantastic talk by Andrew NG about the model-centric approach vs data-centric approach and I find it particularly relevant for this competition.\n\nMy main lesson from this one is: do not try to fix with model architecture something that should be fixed by other elements of ML pipeline. ",
    "1659461": "NeurIPS Data-Centric AI workshop in December 2021 also showed that the data-centric approach is becoming more and more popular; it's fantastic! ",
    "1273174": "I think data-centric approach should be better",
    "1273068": "What are the other elements of ML pipeline?"
  }
}