{
  "id": 154844,
  "title": "Don't flip your labels",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/154844",
  "author_name": "",
  "post_date": "2020-05-30T04:18:23.480224300Z",
  "votes": 1,
  "comment_count": 11,
  "views": 0,
  "content": "<p>Or you'll end up in last place 😅🙃</p>",
  "messages": [
    {
      "id": "867168",
      "postDate": "05/30/2020 04:18:23",
      "content": "<p>Or you'll end up in last place 😅🙃</p>",
      "rawMarkdown": "Or you'll end up in last place 😅🙃",
      "votes": null
    },
    {
      "id": "867197",
      "postDate": "05/30/2020 04:47:38",
      "content": "<p>Foundation for the leaderboard seems pretty strong, eh. \"Everything is fine\" :)</p>\n\n<p>Thanks for the info though! ^_^</p>",
      "rawMarkdown": "Foundation for the leaderboard seems pretty strong, eh. \"Everything is fine\" :)\n\nThanks for the info though! ^_^",
      "votes": null
    },
    {
      "id": "867221",
      "postDate": "05/30/2020 05:28:55",
      "content": "<p>Why did I get downvoted? I’m just pointing out that a flipped AUC (less than 0.5) is just as good as above 0.5. </p>",
      "rawMarkdown": "Why did I get downvoted? I’m just pointing out that a flipped AUC (less than 0.5) is just as good as above 0.5.",
      "votes": null
    },
    {
      "id": "867351",
      "postDate": "05/30/2020 07:33:28",
      "content": "<p>Perhaps <a href=\"/hamzag95\">@hamzag95</a> should have clearly highlighted the <strong>drawbacks</strong> of AUC score.\nFor example, try this: divide all predictions by 10, but AUC score <strong><em>remains the same</em></strong>!</p>\n\n<p>See <a href=\"http://sandeeptata.blogspot.com/2015/04/on-dangers-of-auc.html\">On the dangers of AUC</a></p>\n\n<p>In some situations, AUC score can even be <strong>incoherent</strong> - see <a href=\"https://www.sciencedirect.com/science/article/abs/pii/S0167865512003923\">paper</a></p>",
      "rawMarkdown": "Perhaps @hamzag95 should have clearly highlighted the **drawbacks** of AUC score.\nFor example, try this: divide all predictions by 10, but AUC score ***remains the same***!\n\nSee [On the dangers of AUC](http://sandeeptata.blogspot.com/2015/04/on-dangers-of-auc.html)\n\nIn some situations, AUC score can even be **incoherent** - see [paper](https://www.sciencedirect.com/science/article/abs/pii/S0167865512003923)",
      "votes": null
    },
    {
      "id": "867845",
      "postDate": "05/30/2020 17:10:53",
      "content": "<p>Kaggle is smart about that 1 - AUC == AUC  is 'True' for kaggle LB</p>",
      "rawMarkdown": "Kaggle is smart about that 1 - AUC == AUC  is 'True' for kaggle LB",
      "votes": null
    },
    {
      "id": "868095",
      "postDate": "05/30/2020 23:23:27",
      "content": "<p>AUC has also been criticized because patients don't present to the clinic in normal-diseased pairs - one with melanoma, and another without. </p>",
      "rawMarkdown": "AUC has also been criticized because patients don't present to the clinic in normal-diseased pairs - one with melanoma, and another without.",
      "votes": null
    },
    {
      "id": "868122",
      "postDate": "05/31/2020 00:25:16",
      "content": "<p>Could you elaborate what you mean?</p>",
      "rawMarkdown": "Could you elaborate what you mean?",
      "votes": null
    },
    {
      "id": "868557",
      "postDate": "05/31/2020 10:29:05",
      "content": "<p>Try it for yourself. flip the label probability and submit and you will know. </p>",
      "rawMarkdown": "Try it for yourself. flip the label probability and submit and you will know.",
      "votes": null
    },
    {
      "id": "868627",
      "postDate": "05/31/2020 11:30:07",
      "content": "<p>I did. It’s works as expected, went from 0.23 to 0.77. I thought you meant something else </p>",
      "rawMarkdown": "I did. It’s works as expected, went from 0.23 to 0.77. I thought you meant something else",
      "votes": null
    },
    {
      "id": "868632",
      "postDate": "05/31/2020 11:36:50",
      "content": "<p>Actually it is supposed to adjust automatically. In past, I have experimented that. May be its different in this one. </p>",
      "rawMarkdown": "Actually it is supposed to adjust automatically. In past, I have experimented that. May be its different in this one.",
      "votes": null
    },
    {
      "id": "869137",
      "postDate": "05/31/2020 18:54:27",
      "content": "<p>may be AUC  was flipped to hear that and downvoted you </p>",
      "rawMarkdown": "may be AUC  was flipped to hear that and downvoted you",
      "votes": null
    },
    {
      "id": "869167",
      "postDate": "05/31/2020 19:31:33",
      "content": "<p>hahah i’m gonna assume this is the case </p>",
      "rawMarkdown": "hahah i’m gonna assume this is the case",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 867197,
      "author_name": "sarques",
      "author_url": "",
      "post_date": "05/30/2020 04:47:38",
      "content": "<p>Foundation for the leaderboard seems pretty strong, eh. \"Everything is fine\" :)</p>\n\n<p>Thanks for the info though! ^_^</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 867221,
      "author_name": "hamzag95",
      "author_url": "",
      "post_date": "05/30/2020 05:28:55",
      "content": "<p>Why did I get downvoted? I’m just pointing out that a flipped AUC (less than 0.5) is just as good as above 0.5. </p>",
      "votes": null,
      "replies": [
        {
          "id": 869137,
          "author_name": "phoenix9032",
          "author_url": "",
          "post_date": "05/31/2020 18:54:27",
          "content": "<p>may be AUC  was flipped to hear that and downvoted you </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 869167,
          "author_name": "hamzag95",
          "author_url": "",
          "post_date": "05/31/2020 19:31:33",
          "content": "<p>hahah i’m gonna assume this is the case </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 867351,
      "author_name": "sirishks",
      "author_url": "",
      "post_date": "05/30/2020 07:33:28",
      "content": "<p>Perhaps <a href=\"/hamzag95\">@hamzag95</a> should have clearly highlighted the <strong>drawbacks</strong> of AUC score.\nFor example, try this: divide all predictions by 10, but AUC score <strong><em>remains the same</em></strong>!</p>\n\n<p>See <a href=\"http://sandeeptata.blogspot.com/2015/04/on-dangers-of-auc.html\">On the dangers of AUC</a></p>\n\n<p>In some situations, AUC score can even be <strong>incoherent</strong> - see <a href=\"https://www.sciencedirect.com/science/article/abs/pii/S0167865512003923\">paper</a></p>",
      "votes": null,
      "replies": [
        {
          "id": 868095,
          "author_name": "twopift",
          "author_url": "",
          "post_date": "05/30/2020 23:23:27",
          "content": "<p>AUC has also been criticized because patients don't present to the clinic in normal-diseased pairs - one with melanoma, and another without. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 867845,
      "author_name": "mks2192",
      "author_url": "",
      "post_date": "05/30/2020 17:10:53",
      "content": "<p>Kaggle is smart about that 1 - AUC == AUC  is 'True' for kaggle LB</p>",
      "votes": null,
      "replies": [
        {
          "id": 868122,
          "author_name": "hamzag95",
          "author_url": "",
          "post_date": "05/31/2020 00:25:16",
          "content": "<p>Could you elaborate what you mean?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 868557,
          "author_name": "mks2192",
          "author_url": "",
          "post_date": "05/31/2020 10:29:05",
          "content": "<p>Try it for yourself. flip the label probability and submit and you will know. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 868627,
          "author_name": "hamzag95",
          "author_url": "",
          "post_date": "05/31/2020 11:30:07",
          "content": "<p>I did. It’s works as expected, went from 0.23 to 0.77. I thought you meant something else </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 868632,
          "author_name": "mks2192",
          "author_url": "",
          "post_date": "05/31/2020 11:36:50",
          "content": "<p>Actually it is supposed to adjust automatically. In past, I have experimented that. May be its different in this one. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "867168": "Or you'll end up in last place 😅🙃",
    "867197": "Foundation for the leaderboard seems pretty strong, eh. \"Everything is fine\" :)\n\nThanks for the info though! ^_^",
    "867221": "Why did I get downvoted? I’m just pointing out that a flipped AUC (less than 0.5) is just as good as above 0.5.",
    "867351": "Perhaps @hamzag95 should have clearly highlighted the **drawbacks** of AUC score.\nFor example, try this: divide all predictions by 10, but AUC score ***remains the same***!\n\nSee [On the dangers of AUC](http://sandeeptata.blogspot.com/2015/04/on-dangers-of-auc.html)\n\nIn some situations, AUC score can even be **incoherent** - see [paper](https://www.sciencedirect.com/science/article/abs/pii/S0167865512003923)",
    "867845": "Kaggle is smart about that 1 - AUC == AUC  is 'True' for kaggle LB",
    "868095": "AUC has also been criticized because patients don't present to the clinic in normal-diseased pairs - one with melanoma, and another without.",
    "868122": "Could you elaborate what you mean?",
    "868557": "Try it for yourself. flip the label probability and submit and you will know.",
    "868627": "I did. It’s works as expected, went from 0.23 to 0.77. I thought you meant something else",
    "868632": "Actually it is supposed to adjust automatically. In past, I have experimented that. May be its different in this one.",
    "869137": "may be AUC  was flipped to hear that and downvoted you",
    "869167": "hahah i’m gonna assume this is the case"
  },
  "source": "meta"
}