{
  "id": 186619,
  "title": "Patient Clustering/Unsupervised learning",
  "url": "/competitions/osic-pulmonary-fibrosis-progression/discussion/186619",
  "author_name": "",
  "post_date": "2020-09-25T08:05:04.572903900Z",
  "votes": 5,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Has anyone tried to use unsupervised learning to make cluster of Patient and use it as a feature or one model for each cluster? <br>\nI tried K-Means with K in [2,3,4] but didn't have better results.</p>",
  "messages": [
    {
      "id": "1026309",
      "postDate": "09/25/2020 08:05:04",
      "content": "<p>Has anyone tried to use unsupervised learning to make cluster of Patient and use it as a feature or one model for each cluster? <br>\nI tried K-Means with K in [2,3,4] but didn't have better results.</p>",
      "rawMarkdown": "Has anyone tried to use unsupervised learning to make cluster of Patient and use it as a feature or one model for each cluster? \nI tried K-Means with K in [2,3,4] but didn't have better results.",
      "votes": null
    },
    {
      "id": "1027683",
      "postDate": "09/26/2020 09:48:19",
      "content": "<p>good job ! go ahead</p>",
      "rawMarkdown": "good job ! go ahead",
      "votes": null
    },
    {
      "id": "1028394",
      "postDate": "09/26/2020 20:20:37",
      "content": "<p>I clustered the patients in 3 groups depending of the severity of the lung function decline (fast decline, stable and increase), using <a href=\"https://en.wikipedia.org/wiki/Jenks_natural_breaks_optimization\" target=\"_blank\">jenks</a>.</p>\n<p>The problem of using the cluster index as feature, is that you will also have to provide this feature for the patients in the test set. </p>",
      "rawMarkdown": "I clustered the patients in 3 groups depending of the severity of the lung function decline (fast decline, stable and increase), using [jenks](https://en.wikipedia.org/wiki/Jenks_natural_breaks_optimization).\n\nThe problem of using the cluster index as feature, is that you will also have to provide this feature for the patients in the test set.",
      "votes": null
    },
    {
      "id": "1028443",
      "postDate": "09/26/2020 21:10:59",
      "content": "<p>I didn't know jenks. But yes, I think you can't do it with this method. You need a learning algorithm that will be able to cluster on the test set after training it on the training set. This is why I tried using K-Means.<br>\nThanks for your answer !</p>",
      "rawMarkdown": "I didn't know jenks. But yes, I think you can't do it with this method. You need a learning algorithm that will be able to cluster on the test set after training it on the training set. This is why I tried using K-Means.\nThanks for your answer !",
      "votes": null
    },
    {
      "id": "2381935",
      "postDate": "08/09/2023 13:02:26",
      "content": "<p>Can I see the code and deliverables you've been working on? I want to see your dataset</p>",
      "rawMarkdown": "Can I see the code and deliverables you've been working on? I want to see your dataset",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1028394,
      "author_name": "mavillan",
      "author_url": "",
      "post_date": "09/26/2020 20:20:37",
      "content": "<p>I clustered the patients in 3 groups depending of the severity of the lung function decline (fast decline, stable and increase), using <a href=\"https://en.wikipedia.org/wiki/Jenks_natural_breaks_optimization\" target=\"_blank\">jenks</a>.</p>\n<p>The problem of using the cluster index as feature, is that you will also have to provide this feature for the patients in the test set. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1028443,
          "author_name": "yohannwattiez",
          "author_url": "",
          "post_date": "09/26/2020 21:10:59",
          "content": "<p>I didn't know jenks. But yes, I think you can't do it with this method. You need a learning algorithm that will be able to cluster on the test set after training it on the training set. This is why I tried using K-Means.<br>\nThanks for your answer !</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 2381935,
          "author_name": "ohhyungseok",
          "author_url": "",
          "post_date": "08/09/2023 13:02:26",
          "content": "<p>Can I see the code and deliverables you've been working on? I want to see your dataset</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1027683,
      "author_name": "naim99",
      "author_url": "",
      "post_date": "09/26/2020 09:48:19",
      "content": "<p>good job ! go ahead</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1026309": "Has anyone tried to use unsupervised learning to make cluster of Patient and use it as a feature or one model for each cluster? \nI tried K-Means with K in [2,3,4] but didn't have better results.",
    "1027683": "good job ! go ahead",
    "1028394": "I clustered the patients in 3 groups depending of the severity of the lung function decline (fast decline, stable and increase), using [jenks](https://en.wikipedia.org/wiki/Jenks_natural_breaks_optimization).\n\nThe problem of using the cluster index as feature, is that you will also have to provide this feature for the patients in the test set.",
    "1028443": "I didn't know jenks. But yes, I think you can't do it with this method. You need a learning algorithm that will be able to cluster on the test set after training it on the training set. This is why I tried using K-Means.\nThanks for your answer !",
    "2381935": "Can I see the code and deliverables you've been working on? I want to see your dataset"
  },
  "source": "meta"
}