{
  "id": 169488,
  "title": "CV vs Public Score",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/169488",
  "author_name": "soham mukherjee",
  "post_date": "2020-07-24T03:31:23.456000",
  "votes": 9,
  "comment_count": 8,
  "views": 0,
  "content": "<p>I joined late in this competition! But thanks to the discussion section and public notebooks! Actually I was getting a CV of 0.911 and a score of 0.9469 and from the discussion section, I came to know that there will be a huge shakeup! And after going through most of the public notebooks most of them overfitted too much or tried ensembling! \nAny suggestion about what can be a good CV for this competition?\nin my case:\nEff B6- 512\nCV- 0.911\nPS- 0.9469</p>\n\n<p>Eff B6- 384\nCV- 0.908\nPS- 0.9459</p>\n\n<p>and so on..</p>",
  "messages": [
    {
      "id": 942856,
      "postDate": "2020-07-24T03:31:23.457Z",
      "content": "<p>I joined late in this competition! But thanks to the discussion section and public notebooks! Actually I was getting a CV of 0.911 and a score of 0.9469 and from the discussion section, I came to know that there will be a huge shakeup! And after going through most of the public notebooks most of them overfitted too much or tried ensembling! \nAny suggestion about what can be a good CV for this competition?\nin my case:\nEff B6- 512\nCV- 0.911\nPS- 0.9469</p>\n\n<p>Eff B6- 384\nCV- 0.908\nPS- 0.9459</p>\n\n<p>and so on..</p>",
      "rawMarkdown": "I joined late in this competition! But thanks to the discussion section and public notebooks! Actually I was getting a CV of 0.911 and a score of 0.9469 and from the discussion section, I came to know that there will be a huge shakeup! And after going through most of the public notebooks most of them overfitted too much or tried ensembling! \nAny suggestion about what can be a good CV for this competition?\nin my case:\nEff B6- 512\nCV- 0.911\nPS- 0.9469\n\nEff B6- 384\nCV- 0.908\nPS- 0.9459\n\nand so on..",
      "votes": 8
    },
    {
      "id": 958158,
      "postDate": "2020-08-04T20:00:19.583Z",
      "content": "<p>effnet-b0 on 256x256 images, cv 9195 LB 9211.</p>\n\n<p>some tta and some bagging.  Goal was to have cv-lb gap small.</p>",
      "rawMarkdown": "effnet-b0 on 256x256 images, cv 9195 LB 9211.\n\nsome tta and some bagging.  Goal was to have cv-lb gap small.",
      "votes": 3
    },
    {
      "id": 943495,
      "postDate": "2020-07-24T12:00:15.327Z",
      "content": "<p>Mine is : \nmodel : effnet-b0 (256)\nfold: single-fold (LOOCV)\nCV: 0.9124 ; LB: 0.9191</p>\n\n<p>I'm still at the beginning state, trying new ideas w/ simple model.</p>",
      "rawMarkdown": "Mine is : \nmodel : effnet-b0 (256)\nfold: single-fold (LOOCV)\nCV: 0.9124 ; LB: 0.9191\n\nI'm still at the beginning state, trying new ideas w/ simple model.",
      "votes": 1
    },
    {
      "id": 943281,
      "postDate": "2020-07-24T08:55:33.630Z",
      "content": "<p>What kind of CV do you use?</p>",
      "rawMarkdown": "What kind of CV do you use?",
      "votes": 1,
      "replies": [
        {
          "id": 943382,
          "postDate": "2020-07-24T10:23:42.157Z",
          "content": "<p>Taking avg of 5 folds! Although max CV I got in single-fold - 0.931 </p>",
          "rawMarkdown": "Taking avg of 5 folds! Although max CV I got in single-fold - 0.931 "
        },
        {
          "id": 943670,
          "postDate": "2020-07-24T14:08:34.400Z",
          "content": "<p>I was talking about CV strategy. As for me <a href=\"/cdeotte\">@cdeotte</a>  CV strategy is reliable enough. Maybe you can try it</p>",
          "rawMarkdown": "I was talking about CV strategy. As for me @cdeotte  CV strategy is reliable enough. Maybe you can try it"
        },
        {
          "id": 945274,
          "postDate": "2020-07-25T17:39:04.733Z",
          "content": "<p>with efficientnet you can get .947 cv on some folds and an average cv of .93~ using <a href=\"/cdeotte\">@cdeotte</a> CV strategy. Of course, you need the right setup :P. </p>\n\n<p>Check out the last year's winner's papers and there might be something in there. </p>\n\n<p><a href=\"https://challenge2019.isic-archive.com/leaderboard.html\">https://challenge2019.isic-archive.com/leaderboard.html</a></p>",
          "rawMarkdown": "with efficientnet you can get .947 cv on some folds and an average cv of .93~ using @cdeotte CV strategy. Of course, you need the right setup :P. \n\nCheck out the last year's winner's papers and there might be something in there. \n\nhttps://challenge2019.isic-archive.com/leaderboard.html",
          "votes": 2
        }
      ]
    },
    {
      "id": 943331,
      "postDate": "2020-07-24T09:52:08.467Z",
      "content": "<p>Mine is: \nModel: Eff-B4 (384)\nFold: Single Fold (out of 5 folds)\nCV: 0.926\nLB: 0.9245</p>",
      "rawMarkdown": "Mine is: \nModel: Eff-B4 (384)\nFold: Single Fold (out of 5 folds)\nCV: 0.926\nLB: 0.9245",
      "votes": 2,
      "replies": [
        {
          "id": 943478,
          "postDate": "2020-07-24T11:50:19.490Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 958158,
      "author_name": "CPMP",
      "author_url": "",
      "post_date": "2020-08-04T20:00:19.583000",
      "content": "<p>effnet-b0 on 256x256 images, cv 9195 LB 9211.</p>\n\n<p>some tta and some bagging.  Goal was to have cv-lb gap small.</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 943495,
      "author_name": "godelscat",
      "author_url": "",
      "post_date": "2020-07-24T12:00:15.327000",
      "content": "<p>Mine is : \nmodel : effnet-b0 (256)\nfold: single-fold (LOOCV)\nCV: 0.9124 ; LB: 0.9191</p>\n\n<p>I'm still at the beginning state, trying new ideas w/ simple model.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 943281,
      "author_name": "Volodymyr",
      "author_url": "",
      "post_date": "2020-07-24T08:55:33.630000",
      "content": "<p>What kind of CV do you use?</p>",
      "votes": 1,
      "replies": [
        {
          "id": 943382,
          "author_name": "soham mukherjee",
          "author_url": "",
          "post_date": "2020-07-24T10:23:42.157000",
          "content": "<p>Taking avg of 5 folds! Although max CV I got in single-fold - 0.931 </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 943670,
          "author_name": "Volodymyr",
          "author_url": "",
          "post_date": "2020-07-24T14:08:34.400000",
          "content": "<p>I was talking about CV strategy. As for me <a href=\"/cdeotte\">@cdeotte</a>  CV strategy is reliable enough. Maybe you can try it</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 945274,
          "author_name": "hongy",
          "author_url": "",
          "post_date": "2020-07-25T17:39:04.733000",
          "content": "<p>with efficientnet you can get .947 cv on some folds and an average cv of .93~ using <a href=\"/cdeotte\">@cdeotte</a> CV strategy. Of course, you need the right setup :P. </p>\n\n<p>Check out the last year's winner's papers and there might be something in there. </p>\n\n<p><a href=\"https://challenge2019.isic-archive.com/leaderboard.html\">https://challenge2019.isic-archive.com/leaderboard.html</a></p>",
          "votes": 2,
          "replies": []
        }
      ]
    },
    {
      "id": 943331,
      "author_name": "Dracarys",
      "author_url": "",
      "post_date": "2020-07-24T09:52:08.467000",
      "content": "<p>Mine is: \nModel: Eff-B4 (384)\nFold: Single Fold (out of 5 folds)\nCV: 0.926\nLB: 0.9245</p>",
      "votes": 2,
      "replies": [
        {
          "id": 943478,
          "author_name": "",
          "author_url": "",
          "post_date": "2020-07-24T11:50:19.490000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "942856": "I joined late in this competition! But thanks to the discussion section and public notebooks! Actually I was getting a CV of 0.911 and a score of 0.9469 and from the discussion section, I came to know that there will be a huge shakeup! And after going through most of the public notebooks most of them overfitted too much or tried ensembling! \nAny suggestion about what can be a good CV for this competition?\nin my case:\nEff B6- 512\nCV- 0.911\nPS- 0.9469\n\nEff B6- 384\nCV- 0.908\nPS- 0.9459\n\nand so on..",
    "958158": "effnet-b0 on 256x256 images, cv 9195 LB 9211.\n\nsome tta and some bagging.  Goal was to have cv-lb gap small.",
    "943495": "Mine is : \nmodel : effnet-b0 (256)\nfold: single-fold (LOOCV)\nCV: 0.9124 ; LB: 0.9191\n\nI'm still at the beginning state, trying new ideas w/ simple model.",
    "943281": "What kind of CV do you use?",
    "943331": "Mine is: \nModel: Eff-B4 (384)\nFold: Single Fold (out of 5 folds)\nCV: 0.926\nLB: 0.9245"
  }
}