{
  "id": 238000,
  "title": "Always Trust on your CV",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/238000",
  "author_name": "",
  "post_date": "2021-05-11T00:20:55.957400100Z",
  "votes": 7,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I had submission that achieved <strong>0.947</strong> on the private LB and and 0.915 on public LB. The CV for this model was 0.952 :|. However I chose the wrong models for final submissions. The sad part is that I missed the medals. </p>\n<p>I guess, all these pseudo labeling stuff made us confused. This is also a lessons learned for future competitions. </p>\n<p>Congrats to all winners. </p>",
  "messages": [
    {
      "id": "1301092",
      "postDate": "05/11/2021 00:20:55",
      "content": "<p>I had submission that achieved <strong>0.947</strong> on the private LB and and 0.915 on public LB. The CV for this model was 0.952 :|. However I chose the wrong models for final submissions. The sad part is that I missed the medals. </p>\n<p>I guess, all these pseudo labeling stuff made us confused. This is also a lessons learned for future competitions. </p>\n<p>Congrats to all winners. </p>",
      "rawMarkdown": "I had submission that achieved **0.947** on the private LB and and 0.915 on public LB. The CV for this model was 0.952 :|. However I chose the wrong models for final submissions. The sad part is that I missed the medals. \n\nI guess, all these pseudo labeling stuff made us confused. This is also a lessons learned for future competitions. \n\nCongrats to all winners.",
      "votes": null
    },
    {
      "id": "1301213",
      "postDate": "05/11/2021 01:39:02",
      "content": "<p>Same here.</p>",
      "rawMarkdown": "Same here.",
      "votes": null
    },
    {
      "id": "1301316",
      "postDate": "05/11/2021 02:57:41",
      "content": "<p>Yes, this is the truth.</p>",
      "rawMarkdown": "Yes, this is the truth.",
      "votes": null
    },
    {
      "id": "1301337",
      "postDate": "05/11/2021 03:09:38",
      "content": "<p>The same, PB 0.945 and LB 0.915. The dilemma is we don't know if there exists similar labeling error in PB dataset, so plabel should be one of the two subs. what's your method to achieve 0.947, could you elaborate it?</p>",
      "rawMarkdown": "The same, PB 0.945 and LB 0.915. The dilemma is we don't know if there exists similar labeling error in PB dataset, so plabel should be one of the two subs. what's your method to achieve 0.947, could you elaborate it?",
      "votes": null
    },
    {
      "id": "1301390",
      "postDate": "05/11/2021 04:03:22",
      "content": "<p>I always believe my CV. Maybe next time you should train a excellent model that can fit both LB and PB well</p>",
      "rawMarkdown": "I always believe my CV. Maybe next time you should train a excellent model that can fit both LB and PB well",
      "votes": null
    },
    {
      "id": "1304662",
      "postDate": "05/12/2021 19:02:01",
      "content": "<p>Congratulations on achieving CV 0.952. That is a great accurate model. You will get a medal in your next competition!</p>",
      "rawMarkdown": "Congratulations on achieving CV 0.952. That is a great accurate model. You will get a medal in your next competition!",
      "votes": null
    },
    {
      "id": "1304702",
      "postDate": "05/12/2021 19:29:01",
      "content": "<p>Same here, Local CV score 0.940 which did 0.935 on the LB and almost 0.9437 on the PB. </p>\n<p>I eventually chose some ensembles of models that scored between 0.940 - 0.942 as Local CV and 0.935 - 0.936 on the LB. Bummer that those ensembles did only 0.9419 on the PB.</p>\n<p>But still very happy with the validation strategy I used…a lot of times there was not much difference between Local CV, LB and PB. All within 0.01 … so not bad. there have been worse competitions as far as Local CV, LB and PB differences are concerned.</p>",
      "rawMarkdown": "Same here, Local CV score 0.940 which did 0.935 on the LB and almost 0.9437 on the PB. \n\nI eventually chose some ensembles of models that scored between 0.940 - 0.942 as Local CV and 0.935 - 0.936 on the LB. Bummer that those ensembles did only 0.9419 on the PB.\n\nBut still very happy with the validation strategy I used...a lot of times there was not much difference between Local CV, LB and PB. All within 0.01 ... so not bad. there have been worse competitions as far as Local CV, LB and PB differences are concerned.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1301213,
      "author_name": "erikdali",
      "author_url": "",
      "post_date": "05/11/2021 01:39:02",
      "content": "<p>Same here.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1301316,
      "author_name": "paulsuen",
      "author_url": "",
      "post_date": "05/11/2021 02:57:41",
      "content": "<p>Yes, this is the truth.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1301337,
      "author_name": "plugin1689",
      "author_url": "",
      "post_date": "05/11/2021 03:09:38",
      "content": "<p>The same, PB 0.945 and LB 0.915. The dilemma is we don't know if there exists similar labeling error in PB dataset, so plabel should be one of the two subs. what's your method to achieve 0.947, could you elaborate it?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1301390,
      "author_name": "luciusk",
      "author_url": "",
      "post_date": "05/11/2021 04:03:22",
      "content": "<p>I always believe my CV. Maybe next time you should train a excellent model that can fit both LB and PB well</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1304662,
      "author_name": "cdeotte",
      "author_url": "",
      "post_date": "05/12/2021 19:02:01",
      "content": "<p>Congratulations on achieving CV 0.952. That is a great accurate model. You will get a medal in your next competition!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1304702,
      "author_name": "rsmits",
      "author_url": "",
      "post_date": "05/12/2021 19:29:01",
      "content": "<p>Same here, Local CV score 0.940 which did 0.935 on the LB and almost 0.9437 on the PB. </p>\n<p>I eventually chose some ensembles of models that scored between 0.940 - 0.942 as Local CV and 0.935 - 0.936 on the LB. Bummer that those ensembles did only 0.9419 on the PB.</p>\n<p>But still very happy with the validation strategy I used…a lot of times there was not much difference between Local CV, LB and PB. All within 0.01 … so not bad. there have been worse competitions as far as Local CV, LB and PB differences are concerned.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1301092": "I had submission that achieved **0.947** on the private LB and and 0.915 on public LB. The CV for this model was 0.952 :|. However I chose the wrong models for final submissions. The sad part is that I missed the medals. \n\nI guess, all these pseudo labeling stuff made us confused. This is also a lessons learned for future competitions. \n\nCongrats to all winners.",
    "1301213": "Same here.",
    "1301316": "Yes, this is the truth.",
    "1301337": "The same, PB 0.945 and LB 0.915. The dilemma is we don't know if there exists similar labeling error in PB dataset, so plabel should be one of the two subs. what's your method to achieve 0.947, could you elaborate it?",
    "1301390": "I always believe my CV. Maybe next time you should train a excellent model that can fit both LB and PB well",
    "1304662": "Congratulations on achieving CV 0.952. That is a great accurate model. You will get a medal in your next competition!",
    "1304702": "Same here, Local CV score 0.940 which did 0.935 on the LB and almost 0.9437 on the PB. \n\nI eventually chose some ensembles of models that scored between 0.940 - 0.942 as Local CV and 0.935 - 0.936 on the LB. Bummer that those ensembles did only 0.9419 on the PB.\n\nBut still very happy with the validation strategy I used...a lot of times there was not much difference between Local CV, LB and PB. All within 0.01 ... so not bad. there have been worse competitions as far as Local CV, LB and PB differences are concerned."
  },
  "source": "meta"
}