{
  "id": 683833,
  "title": "Any idea to deal with the lizard dataset ?",
  "url": "/competitions/animal-clef-2026/discussion/683833",
  "author_name": "",
  "post_date": "2026-03-22T13:32:43.592712100Z",
  "votes": null,
  "comment_count": 2,
  "views": 0,
  "content": "<p>As we all know, the training data of lizards is not provided to us. So I just don't get how are we supposed to cluster them, cause even if we cluster the testing data how would we know the number of clusters in which we need to group the testing species ?</p>",
  "messages": [
    {
      "id": "3426376",
      "postDate": "03/22/2026 13:32:43",
      "content": "<p>As we all know, the training data of lizards is not provided to us. So I just don't get how are we supposed to cluster them, cause even if we cluster the testing data how would we know the number of clusters in which we need to group the testing species ?</p>",
      "rawMarkdown": "As we all know, the training data of lizards is not provided to us. So I just don't get how are we supposed to cluster them, cause even if we cluster the testing data how would we know the number of clusters in which we need to group the testing species ?",
      "votes": null
    },
    {
      "id": "3426395",
      "postDate": "03/22/2026 14:07:17",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/kushubhai\" target=\"_blank\">@kushubhai</a> ,</p>\n<p>All species are in the same situation here. The training data does not contain all identities present in the test data.</p>\n<p>This is central to the ecological scenarios behind the competition. In real wildlife monitoring, no species comes with a complete sample set of all individuals. Observations are partial, uneven, and incomplete. Some individuals are seen many times, some only once, and others not at all. In most cases, the total number of individuals is simply unknown beforehand.</p>\n<p>The competition is designed to reflect that reality. The goal is not to rely on a fixed, complete list of known identities, but to develop methods that can handle incomplete sampling and an unknown number of individuals.</p>\n<p>Best,\nLukas</p>",
      "rawMarkdown": "Hi @kushubhai ,\n\nAll species are in the same situation here. The training data does not contain all identities present in the test data.\n\nThis is central to the ecological scenarios behind the competition. In real wildlife monitoring, no species comes with a complete sample set of all individuals. Observations are partial, uneven, and incomplete. Some individuals are seen many times, some only once, and others not at all. In most cases, the total number of individuals is simply unknown beforehand.\n\nThe competition is designed to reflect that reality. The goal is not to rely on a fixed, complete list of known identities, but to develop methods that can handle incomplete sampling and an unknown number of individuals.\n\nBest,\nLukas",
      "votes": null
    },
    {
      "id": "3426522",
      "postDate": "03/22/2026 18:28:05",
      "content": "<p>Alright Sir, thanks 👍\nWill keep that in mind 👍</p>",
      "rawMarkdown": "Alright Sir, thanks 👍\nWill keep that in mind 👍",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3426395,
      "author_name": "picekl",
      "author_url": "",
      "post_date": "03/22/2026 14:07:17",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/kushubhai\" target=\"_blank\">@kushubhai</a> ,</p>\n<p>All species are in the same situation here. The training data does not contain all identities present in the test data.</p>\n<p>This is central to the ecological scenarios behind the competition. In real wildlife monitoring, no species comes with a complete sample set of all individuals. Observations are partial, uneven, and incomplete. Some individuals are seen many times, some only once, and others not at all. In most cases, the total number of individuals is simply unknown beforehand.</p>\n<p>The competition is designed to reflect that reality. The goal is not to rely on a fixed, complete list of known identities, but to develop methods that can handle incomplete sampling and an unknown number of individuals.</p>\n<p>Best,\nLukas</p>",
      "votes": null,
      "replies": [
        {
          "id": 3426522,
          "author_name": "kushubhai",
          "author_url": "",
          "post_date": "03/22/2026 18:28:05",
          "content": "<p>Alright Sir, thanks 👍\nWill keep that in mind 👍</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3426376": "As we all know, the training data of lizards is not provided to us. So I just don't get how are we supposed to cluster them, cause even if we cluster the testing data how would we know the number of clusters in which we need to group the testing species ?",
    "3426395": "Hi @kushubhai ,\n\nAll species are in the same situation here. The training data does not contain all identities present in the test data.\n\nThis is central to the ecological scenarios behind the competition. In real wildlife monitoring, no species comes with a complete sample set of all individuals. Observations are partial, uneven, and incomplete. Some individuals are seen many times, some only once, and others not at all. In most cases, the total number of individuals is simply unknown beforehand.\n\nThe competition is designed to reflect that reality. The goal is not to rely on a fixed, complete list of known identities, but to develop methods that can handle incomplete sampling and an unknown number of individuals.\n\nBest,\nLukas",
    "3426522": "Alright Sir, thanks 👍\nWill keep that in mind 👍"
  },
  "source": "meta"
}