{
  "id": 216241,
  "title": "Is the sample test image class 2 or class 4?",
  "url": "/competitions/cassava-leaf-disease-classification/discussion/216241",
  "author_name": "",
  "post_date": "2021-02-02T05:23:51.294525100Z",
  "votes": 10,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Most of the models I trained were not very confident in the sample image. </p>\n<p>I trained myself today with the train set (LOL) and find that model.ME also not very sure about that.<br>\nIs there any expert knows that?  Very curious.</p>\n<p>Top 8 ensembles of my team votes 50/50 on that.</p>",
  "messages": [
    {
      "id": "1181759",
      "postDate": "02/02/2021 05:23:51",
      "content": "<p>Most of the models I trained were not very confident in the sample image. </p>\n<p>I trained myself today with the train set (LOL) and find that model.ME also not very sure about that.<br>\nIs there any expert knows that?  Very curious.</p>\n<p>Top 8 ensembles of my team votes 50/50 on that.</p>",
      "rawMarkdown": "Most of the models I trained were not very confident in the sample image. \n\nI trained myself today with the train set (LOL) and find that model.ME also not very sure about that.\nIs there any expert knows that?  Very curious.\n\nTop 8 ensembles of my team votes 50/50 on that.",
      "votes": null
    },
    {
      "id": "1182431",
      "postDate": "02/02/2021 13:00:54",
      "content": "<p>I tried many models and it was almost class 4. But it was class 2 barely. When I get class 2, I did 4-fold TTA and bagging ensemble 4 different models. I think it effects by ensemble… I'm also curious about it :)</p>",
      "rawMarkdown": "I tried many models and it was almost class 4. But it was class 2 barely. When I get class 2, I did 4-fold TTA and bagging ensemble 4 different models. I think it effects by ensemble... I'm also curious about it :)",
      "votes": null
    },
    {
      "id": "1183989",
      "postDate": "02/03/2021 10:40:15",
      "content": "<p>I noticed, when my model is very confident about class 4, it scores less on the LB, however when the model scores higher on class 2, it improves the score on the test set. As we can see from the validation set, many times the class 4 belongs to the cluster representing the class 2. I am asking myself the same question, maybe that's the case?</p>\n<p><a href=\"https://ibb.co/72wXjyx\"><img src=\"https://i.ibb.co/pJtxbX6/tsne.png\" alt=\"tsne\"></a><br><a target=\"_blank\" href=\"https://imgbb.com/\">image upload</a><br></p>",
      "rawMarkdown": "I noticed, when my model is very confident about class 4, it scores less on the LB, however when the model scores higher on class 2, it improves the score on the test set. As we can see from the validation set, many times the class 4 belongs to the cluster representing the class 2. I am asking myself the same question, maybe that's the case?\n\n<a href=\"https://ibb.co/72wXjyx\"><img src=\"https://i.ibb.co/pJtxbX6/tsne.png\" alt=\"tsne\" border=\"0\"></a><br /><a target='_blank' href='https://imgbb.com/'>image upload</a><br />",
      "votes": null
    },
    {
      "id": "1184340",
      "postDate": "02/03/2021 13:50:33",
      "content": "<p>I tried few models and mostly it is class 4 and rarely it is class 2. What is your highest scoring individual model predict - <br>\n2 or 4 ?</p>",
      "rawMarkdown": "I tried few models and mostly it is class 4 and rarely it is class 2. What is your highest scoring individual model predict - \n2 or 4 ?",
      "votes": null
    },
    {
      "id": "1185133",
      "postDate": "02/04/2021 00:53:33",
      "content": "<p>Highest scoring, 0.896 single model, EffNetB4 noisy student. What kind of model did you use, and methods?</p>",
      "rawMarkdown": "Highest scoring, 0.896 single model, EffNetB4 noisy student. What kind of model did you use, and methods?",
      "votes": null
    },
    {
      "id": "1185656",
      "postDate": "02/04/2021 09:11:45",
      "content": "<p>I have used same model but I have  also used old dataset. If you want to try, the details of dataset is present <a href=\"https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215490\" target=\"_blank\">here</a></p>",
      "rawMarkdown": "I have used same model but I have  also used old dataset. If you want to try, the details of dataset is present [here](https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215490)",
      "votes": null
    },
    {
      "id": "1185764",
      "postDate": "02/04/2021 11:00:15",
      "content": "<p>Thank you mate, <br>\nCheers.</p>",
      "rawMarkdown": "Thank you mate, \nCheers.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1182431,
      "author_name": "vkehfdl1",
      "author_url": "",
      "post_date": "02/02/2021 13:00:54",
      "content": "<p>I tried many models and it was almost class 4. But it was class 2 barely. When I get class 2, I did 4-fold TTA and bagging ensemble 4 different models. I think it effects by ensemble… I'm also curious about it :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1183989,
      "author_name": "marjan1111",
      "author_url": "",
      "post_date": "02/03/2021 10:40:15",
      "content": "<p>I noticed, when my model is very confident about class 4, it scores less on the LB, however when the model scores higher on class 2, it improves the score on the test set. As we can see from the validation set, many times the class 4 belongs to the cluster representing the class 2. I am asking myself the same question, maybe that's the case?</p>\n<p><a href=\"https://ibb.co/72wXjyx\"><img src=\"https://i.ibb.co/pJtxbX6/tsne.png\" alt=\"tsne\"></a><br><a target=\"_blank\" href=\"https://imgbb.com/\">image upload</a><br></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1184340,
      "author_name": "vickygoyal",
      "author_url": "",
      "post_date": "02/03/2021 13:50:33",
      "content": "<p>I tried few models and mostly it is class 4 and rarely it is class 2. What is your highest scoring individual model predict - <br>\n2 or 4 ?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1185133,
          "author_name": "marjan1111",
          "author_url": "",
          "post_date": "02/04/2021 00:53:33",
          "content": "<p>Highest scoring, 0.896 single model, EffNetB4 noisy student. What kind of model did you use, and methods?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1185656,
          "author_name": "vickygoyal",
          "author_url": "",
          "post_date": "02/04/2021 09:11:45",
          "content": "<p>I have used same model but I have  also used old dataset. If you want to try, the details of dataset is present <a href=\"https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215490\" target=\"_blank\">here</a></p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1185764,
          "author_name": "marjan1111",
          "author_url": "",
          "post_date": "02/04/2021 11:00:15",
          "content": "<p>Thank you mate, <br>\nCheers.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1181759": "Most of the models I trained were not very confident in the sample image. \n\nI trained myself today with the train set (LOL) and find that model.ME also not very sure about that.\nIs there any expert knows that?  Very curious.\n\nTop 8 ensembles of my team votes 50/50 on that.",
    "1182431": "I tried many models and it was almost class 4. But it was class 2 barely. When I get class 2, I did 4-fold TTA and bagging ensemble 4 different models. I think it effects by ensemble... I'm also curious about it :)",
    "1183989": "I noticed, when my model is very confident about class 4, it scores less on the LB, however when the model scores higher on class 2, it improves the score on the test set. As we can see from the validation set, many times the class 4 belongs to the cluster representing the class 2. I am asking myself the same question, maybe that's the case?\n\n<a href=\"https://ibb.co/72wXjyx\"><img src=\"https://i.ibb.co/pJtxbX6/tsne.png\" alt=\"tsne\" border=\"0\"></a><br /><a target='_blank' href='https://imgbb.com/'>image upload</a><br />",
    "1184340": "I tried few models and mostly it is class 4 and rarely it is class 2. What is your highest scoring individual model predict - \n2 or 4 ?",
    "1185133": "Highest scoring, 0.896 single model, EffNetB4 noisy student. What kind of model did you use, and methods?",
    "1185656": "I have used same model but I have  also used old dataset. If you want to try, the details of dataset is present [here](https://www.kaggle.com/c/cassava-leaf-disease-classification/discussion/215490)",
    "1185764": "Thank you mate, \nCheers."
  },
  "source": "meta"
}