{
  "id": 307125,
  "title": "Why standard classification methods won't work...",
  "url": "/competitions/happy-whale-and-dolphin/discussion/307125",
  "author_name": "",
  "post_date": "2022-02-12T17:58:44.837052800Z",
  "votes": 9,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Out of 15587 unique individual _id, more than 9000 of them have just 1 sample, so standard CNNs are not expected to work as they require a large number of samples per label. </p>",
  "messages": [
    {
      "id": "1687281",
      "postDate": "02/12/2022 17:58:44",
      "content": "<p>Out of 15587 unique individual _id, more than 9000 of them have just 1 sample, so standard CNNs are not expected to work as they require a large number of samples per label. </p>",
      "rawMarkdown": "Out of 15587 unique individual _id, more than 9000 of them have just 1 sample, so standard CNNs are not expected to work as they require a large number of samples per label.",
      "votes": null
    },
    {
      "id": "1689184",
      "postDate": "02/14/2022 04:31:55",
      "content": "<p>and having 15587 classes is also not a good idea :)</p>",
      "rawMarkdown": "and having 15587 classes is also not a good idea :)",
      "votes": null
    },
    {
      "id": "1689257",
      "postDate": "02/14/2022 05:44:41",
      "content": "<p>yeah, a siamese network would be a better choice for this type of classification</p>",
      "rawMarkdown": "yeah, a siamese network would be a better choice for this type of classification",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1689184,
      "author_name": "bsridatta",
      "author_url": "",
      "post_date": "02/14/2022 04:31:55",
      "content": "<p>and having 15587 classes is also not a good idea :)</p>",
      "votes": null,
      "replies": [
        {
          "id": 1689257,
          "author_name": "mushfirat",
          "author_url": "",
          "post_date": "02/14/2022 05:44:41",
          "content": "<p>yeah, a siamese network would be a better choice for this type of classification</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1687281": "Out of 15587 unique individual _id, more than 9000 of them have just 1 sample, so standard CNNs are not expected to work as they require a large number of samples per label.",
    "1689184": "and having 15587 classes is also not a good idea :)",
    "1689257": "yeah, a siamese network would be a better choice for this type of classification"
  },
  "source": "meta"
}