{
  "id": 206720,
  "title": "CutMix results in abysmal submission scores",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/206720",
  "author_name": "",
  "post_date": "2020-12-26T08:12:58.242347Z",
  "votes": null,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I implemented CutMix from this kernel: <a href=\"https://www.kaggle.com/ipythonx/tf-cassava-leaf-cutmix-mixup-fmix-tta-ensemble\" target=\"_blank\">https://www.kaggle.com/ipythonx/tf-cassava-leaf-cutmix-mixup-fmix-tta-ensemble</a></p>\n<p>and got a submission score of 0.048, even though my validation data had 89+% accuracy. Anyone has a clue why?</p>",
  "messages": [
    {
      "id": "1127068",
      "postDate": "12/26/2020 08:12:58",
      "content": "<p>I implemented CutMix from this kernel: <a href=\"https://www.kaggle.com/ipythonx/tf-cassava-leaf-cutmix-mixup-fmix-tta-ensemble\" target=\"_blank\">https://www.kaggle.com/ipythonx/tf-cassava-leaf-cutmix-mixup-fmix-tta-ensemble</a></p>\n<p>and got a submission score of 0.048, even though my validation data had 89+% accuracy. Anyone has a clue why?</p>",
      "rawMarkdown": "I implemented CutMix from this kernel: https://www.kaggle.com/ipythonx/tf-cassava-leaf-cutmix-mixup-fmix-tta-ensemble\n\nand got a submission score of 0.048, even though my validation data had 89+% accuracy. Anyone has a clue why?",
      "votes": null
    },
    {
      "id": "1127099",
      "postDate": "12/26/2020 08:30:53",
      "content": "<p>I also enjoyed that exact score.  Check your finished kernel - you had a non fatal error as most likely root cause.</p>",
      "rawMarkdown": "I also enjoyed that exact score.  Check your finished kernel - you had a non fatal error as most likely root cause.",
      "votes": null
    },
    {
      "id": "1127104",
      "postDate": "12/26/2020 08:40:21",
      "content": "<p>Do you mind elaborating? What was the error in your case?</p>",
      "rawMarkdown": "Do you mind elaborating? What was the error in your case?",
      "votes": null
    },
    {
      "id": "1127823",
      "postDate": "12/26/2020 22:55:18",
      "content": "<p>IIRC there is 4.8% 0's in the public test set. Predicting all zeroes will give 0.048</p>",
      "rawMarkdown": "IIRC there is 4.8% 0's in the public test set. Predicting all zeroes will give 0.048",
      "votes": null
    },
    {
      "id": "1127958",
      "postDate": "12/27/2020 04:20:16",
      "content": "<p>Junyl - when I scrolled thru my code after the submission I found an error in cell that was critical to the model (think it was weights file not found) - the script kept on going and a \"bad\" model was able to generate a submission - thanks to <a href=\"url\" target=\"_blank\">ilovescience</a> I now know that I predicted all zero's.</p>",
      "rawMarkdown": "Junyl - when I scrolled thru my code after the submission I found an error in cell that was critical to the model (think it was weights file not found) - the script kept on going and a \"bad\" model was able to generate a submission - thanks to [ilovescience](url) I now know that I predicted all zero's.",
      "votes": null
    },
    {
      "id": "1128761",
      "postDate": "12/27/2020 17:53:16",
      "content": "<p>Thanks a lot!</p>",
      "rawMarkdown": "Thanks a lot!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1127099,
      "author_name": "pcjimmmy",
      "author_url": "",
      "post_date": "12/26/2020 08:30:53",
      "content": "<p>I also enjoyed that exact score.  Check your finished kernel - you had a non fatal error as most likely root cause.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1127104,
          "author_name": "junyingsg",
          "author_url": "",
          "post_date": "12/26/2020 08:40:21",
          "content": "<p>Do you mind elaborating? What was the error in your case?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1127823,
          "author_name": "tanlikesmath",
          "author_url": "",
          "post_date": "12/26/2020 22:55:18",
          "content": "<p>IIRC there is 4.8% 0's in the public test set. Predicting all zeroes will give 0.048</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1127958,
          "author_name": "pcjimmmy",
          "author_url": "",
          "post_date": "12/27/2020 04:20:16",
          "content": "<p>Junyl - when I scrolled thru my code after the submission I found an error in cell that was critical to the model (think it was weights file not found) - the script kept on going and a \"bad\" model was able to generate a submission - thanks to <a href=\"url\" target=\"_blank\">ilovescience</a> I now know that I predicted all zero's.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1128761,
          "author_name": "junyingsg",
          "author_url": "",
          "post_date": "12/27/2020 17:53:16",
          "content": "<p>Thanks a lot!</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1127068": "I implemented CutMix from this kernel: https://www.kaggle.com/ipythonx/tf-cassava-leaf-cutmix-mixup-fmix-tta-ensemble\n\nand got a submission score of 0.048, even though my validation data had 89+% accuracy. Anyone has a clue why?",
    "1127099": "I also enjoyed that exact score.  Check your finished kernel - you had a non fatal error as most likely root cause.",
    "1127104": "Do you mind elaborating? What was the error in your case?",
    "1127823": "IIRC there is 4.8% 0's in the public test set. Predicting all zeroes will give 0.048",
    "1127958": "Junyl - when I scrolled thru my code after the submission I found an error in cell that was critical to the model (think it was weights file not found) - the script kept on going and a \"bad\" model was able to generate a submission - thanks to [ilovescience](url) I now know that I predicted all zero's.",
    "1128761": "Thanks a lot!"
  },
  "source": "meta"
}