{
  "id": 339955,
  "title": "Good CV score?",
  "url": "/competitions/mayo-clinic-strip-ai/discussion/339955",
  "author_name": "",
  "post_date": "2022-07-26T22:25:54.669000",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I've tried many different models, but I still can't get a CV score better than the optimal \"sample submission\". It would be interesting to know: has anyone got meaningful CV score (logloss&lt;0.56 for an unbalanced dataset and logloss&lt;0.69 for a balanced)?</p>",
  "messages": [
    {
      "id": 1875326,
      "postDate": "2022-07-28T23:05:49.873Z",
      "content": "<p>I got 0.61088 val loss using resnet50 in unbalanced datasets.</p>",
      "rawMarkdown": "I got 0.61088 val loss using resnet50 in unbalanced datasets.",
      "votes": 1
    },
    {
      "id": 1873424,
      "postDate": "2022-07-27T15:48:27.683Z",
      "content": "<p>I opened a discussion about how the objective of the competition doesnt seem achievable with this dataset. </p>\n<p>This model is probably The Model for this competition: <a href=\"https://www.youtube.com/watch?v=cABkB1J-GTA\" target=\"_blank\">https://www.youtube.com/watch?v=cABkB1J-GTA</a></p>\n<p>But look at the images they use, the blood sample occupies the whole image. The total pixels are on pair with what we use but key difference on how much pixels are actually useful data.</p>",
      "rawMarkdown": "I opened a discussion about how the objective of the competition doesnt seem achievable with this dataset. \n\nThis model is probably The Model for this competition: https://www.youtube.com/watch?v=cABkB1J-GTA\n\nBut look at the images they use, the blood sample occupies the whole image. The total pixels are on pair with what we use but key difference on how much pixels are actually useful data.",
      "votes": 1
    },
    {
      "id": 1872387,
      "postDate": "2022-07-26T22:25:54.670Z",
      "content": "<p>I've tried many different models, but I still can't get a CV score better than the optimal \"sample submission\". It would be interesting to know: has anyone got meaningful CV score (logloss&lt;0.56 for an unbalanced dataset and logloss&lt;0.69 for a balanced)?</p>",
      "rawMarkdown": "I've tried many different models, but I still can't get a CV score better than the optimal \"sample submission\". It would be interesting to know: has anyone got meaningful CV score (logloss<0.56 for an unbalanced dataset and logloss<0.69 for a balanced)?",
      "votes": 2
    },
    {
      "id": 1939830,
      "postDate": "2022-09-15T03:48:30.983Z",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/realneuralnetwork\" target=\"_blank\">@realneuralnetwork</a> ， How did you implement the Logloss?</p>",
      "rawMarkdown": "Hi @realneuralnetwork ， How did you implement the Logloss?"
    }
  ],
  "comments": [
    {
      "id": 1875326,
      "author_name": "Dominador Ramos Jr",
      "author_url": "",
      "post_date": "2022-07-28T23:05:49.873000",
      "content": "<p>I got 0.61088 val loss using resnet50 in unbalanced datasets.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1873424,
      "author_name": "Iuryck Santos",
      "author_url": "",
      "post_date": "2022-07-27T15:48:27.683000",
      "content": "<p>I opened a discussion about how the objective of the competition doesnt seem achievable with this dataset. </p>\n<p>This model is probably The Model for this competition: <a href=\"https://www.youtube.com/watch?v=cABkB1J-GTA\" target=\"_blank\">https://www.youtube.com/watch?v=cABkB1J-GTA</a></p>\n<p>But look at the images they use, the blood sample occupies the whole image. The total pixels are on pair with what we use but key difference on how much pixels are actually useful data.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1939830,
      "author_name": "ForcewithMe",
      "author_url": "",
      "post_date": "2022-09-15T03:48:30.983000",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/realneuralnetwork\" target=\"_blank\">@realneuralnetwork</a> ， How did you implement the Logloss?</p>",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1875326": "I got 0.61088 val loss using resnet50 in unbalanced datasets.",
    "1873424": "I opened a discussion about how the objective of the competition doesnt seem achievable with this dataset. \n\nThis model is probably The Model for this competition: https://www.youtube.com/watch?v=cABkB1J-GTA\n\nBut look at the images they use, the blood sample occupies the whole image. The total pixels are on pair with what we use but key difference on how much pixels are actually useful data.",
    "1872387": "I've tried many different models, but I still can't get a CV score better than the optimal \"sample submission\". It would be interesting to know: has anyone got meaningful CV score (logloss<0.56 for an unbalanced dataset and logloss<0.69 for a balanced)?",
    "1939830": "Hi @realneuralnetwork ， How did you implement the Logloss?"
  }
}