{
  "id": 221918,
  "title": "Multi Class & Multi Label Classficiation ",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/221918",
  "author_name": "",
  "post_date": "2021-02-24T13:38:14.569910300Z",
  "votes": 11,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Hello Kaggler, i m new here, <br>\ni've question about this competition, can we consider it as Multi-Class classification with multi level ?  </p>\n<p>Because we've labels [CVC ETT NGT ] with 3 classes for ETT CVC --&gt; [Normal Borderline Abnormal ] and 4 classes for NGT [Normal Borderline Abnormal  Incompletely Imaged ] </p>\n<p>Thanks a lot and sorry for the inconvenience </p>",
  "messages": [
    {
      "id": "1216788",
      "postDate": "02/24/2021 13:38:14",
      "content": "<p>Hello Kaggler, i m new here, <br>\ni've question about this competition, can we consider it as Multi-Class classification with multi level ?  </p>\n<p>Because we've labels [CVC ETT NGT ] with 3 classes for ETT CVC --&gt; [Normal Borderline Abnormal ] and 4 classes for NGT [Normal Borderline Abnormal  Incompletely Imaged ] </p>\n<p>Thanks a lot and sorry for the inconvenience </p>",
      "rawMarkdown": "Hello Kaggler, i m new here, \ni've question about this competition, can we consider it as Multi-Class classification with multi level ?  \n\nBecause we've labels [CVC ETT NGT ] with 3 classes for ETT CVC --> [Normal Borderline Abnormal ] and 4 classes for NGT [Normal Borderline Abnormal  Incompletely Imaged ] \n\nThanks a lot and sorry for the inconvenience",
      "votes": null
    },
    {
      "id": "1216898",
      "postDate": "02/24/2021 15:36:56",
      "content": "<p>Hello!</p>\n<p>The most common solution unless you decide to go for catheter detection is to consider this task as multi-label i.e. work with 11 <code>sigmoid</code> neurons on top of your CNN and <code>binary_crossentropy</code> as the loss function.</p>",
      "rawMarkdown": "Hello!\n\nThe most common solution unless you decide to go for catheter detection is to consider this task as multi-label i.e. work with 11 `sigmoid` neurons on top of your CNN and `binary_crossentropy` as the loss function.",
      "votes": null
    },
    {
      "id": "1222478",
      "postDate": "03/01/2021 20:26:17",
      "content": "<p>Thanx for sharing </p>",
      "rawMarkdown": "Thanx for sharing",
      "votes": null
    },
    {
      "id": "1222527",
      "postDate": "03/01/2021 21:35:22",
      "content": "<p>Yes, I have tried composing the problem as two stages, one as classification, is it any of the 4 catheter types, then the further classification of the catheters into their condition. Was not able to see any performance benefit unfortunately. Maybe others have been able to get it to work. </p>",
      "rawMarkdown": "Yes, I have tried composing the problem as two stages, one as classification, is it any of the 4 catheter types, then the further classification of the catheters into their condition. Was not able to see any performance benefit unfortunately. Maybe others have been able to get it to work.",
      "votes": null
    },
    {
      "id": "1230280",
      "postDate": "03/08/2021 01:07:02",
      "content": "<p>Thanks a lot for your reply </p>",
      "rawMarkdown": "Thanks a lot for your reply",
      "votes": null
    },
    {
      "id": "1230281",
      "postDate": "03/08/2021 01:09:18",
      "content": "<p>After a lot of reading, there is some interestings discussion about strategies </p>",
      "rawMarkdown": "After a lot of reading, there is some interestings discussion about strategies",
      "votes": null
    },
    {
      "id": "1230282",
      "postDate": "03/08/2021 01:10:57",
      "content": "<p>Hello there, <br>\nsorry for the late, i would like to thank u for your answer, <br>\nI found a lot of interestings discussion about the strategies and a good kernels </p>",
      "rawMarkdown": "Hello there, \nsorry for the late, i would like to thank u for your answer, \nI found a lot of interestings discussion about the strategies and a good kernels",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1216898,
      "author_name": "nickuzmenkov",
      "author_url": "",
      "post_date": "02/24/2021 15:36:56",
      "content": "<p>Hello!</p>\n<p>The most common solution unless you decide to go for catheter detection is to consider this task as multi-label i.e. work with 11 <code>sigmoid</code> neurons on top of your CNN and <code>binary_crossentropy</code> as the loss function.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1230280,
          "author_name": "salimkhazem",
          "author_url": "",
          "post_date": "03/08/2021 01:07:02",
          "content": "<p>Thanks a lot for your reply </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1222478,
      "author_name": "anisxk",
      "author_url": "",
      "post_date": "03/01/2021 20:26:17",
      "content": "<p>Thanx for sharing </p>",
      "votes": null,
      "replies": [
        {
          "id": 1230281,
          "author_name": "salimkhazem",
          "author_url": "",
          "post_date": "03/08/2021 01:09:18",
          "content": "<p>After a lot of reading, there is some interestings discussion about strategies </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1222527,
      "author_name": "ryches",
      "author_url": "",
      "post_date": "03/01/2021 21:35:22",
      "content": "<p>Yes, I have tried composing the problem as two stages, one as classification, is it any of the 4 catheter types, then the further classification of the catheters into their condition. Was not able to see any performance benefit unfortunately. Maybe others have been able to get it to work. </p>",
      "votes": null,
      "replies": [
        {
          "id": 1230282,
          "author_name": "salimkhazem",
          "author_url": "",
          "post_date": "03/08/2021 01:10:57",
          "content": "<p>Hello there, <br>\nsorry for the late, i would like to thank u for your answer, <br>\nI found a lot of interestings discussion about the strategies and a good kernels </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1216788": "Hello Kaggler, i m new here, \ni've question about this competition, can we consider it as Multi-Class classification with multi level ?  \n\nBecause we've labels [CVC ETT NGT ] with 3 classes for ETT CVC --> [Normal Borderline Abnormal ] and 4 classes for NGT [Normal Borderline Abnormal  Incompletely Imaged ] \n\nThanks a lot and sorry for the inconvenience",
    "1216898": "Hello!\n\nThe most common solution unless you decide to go for catheter detection is to consider this task as multi-label i.e. work with 11 `sigmoid` neurons on top of your CNN and `binary_crossentropy` as the loss function.",
    "1222478": "Thanx for sharing",
    "1222527": "Yes, I have tried composing the problem as two stages, one as classification, is it any of the 4 catheter types, then the further classification of the catheters into their condition. Was not able to see any performance benefit unfortunately. Maybe others have been able to get it to work.",
    "1230280": "Thanks a lot for your reply",
    "1230281": "After a lot of reading, there is some interestings discussion about strategies",
    "1230282": "Hello there, \nsorry for the late, i would like to thank u for your answer, \nI found a lot of interestings discussion about the strategies and a good kernels"
  },
  "source": "meta"
}