{
  "id": 100583,
  "title": "Too low score!!!!",
  "url": "/competitions/aptos2019-blindness-detection/discussion/100583",
  "author_name": "",
  "post_date": "2019-07-19T11:17:04.390974600Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>I have got 0.937 kappa score on train set, however on submission i got 0.014\nany idea why this is happening?\ni have used resent 152 model</p>",
  "messages": [
    {
      "id": "579884",
      "postDate": "07/19/2019 11:17:04",
      "content": "<p>I have got 0.937 kappa score on train set, however on submission i got 0.014\nany idea why this is happening?\ni have used resent 152 model</p>",
      "rawMarkdown": "I have got 0.937 kappa score on train set, however on submission i got 0.014\nany idea why this is happening?\ni have used resent 152 model",
      "votes": null
    },
    {
      "id": "579947",
      "postDate": "07/19/2019 12:31:23",
      "content": "<p>probably overfitting: You train a big model on a small dataset -&gt; big model just memorize all train images -&gt; bad on new images but really good at the train images</p>\n\n<p>Do you have used any validation data? Then you could see how well the models works on new images. \nIf not, try using a smaller model like ResNet-18 or 34 with some validation data.\nIf you are using validation data and the score is pretty good, you maybe generate a wrong submission csv</p>",
      "rawMarkdown": "probably overfitting: You train a big model on a small dataset -&gt; big model just memorize all train images -&gt; bad on new images but really good at the train images\n\nDo you have used any validation data? Then you could see how well the models works on new images. \nIf not, try using a smaller model like ResNet-18 or 34 with some validation data.\nIf you are using validation data and the score is pretty good, you maybe generate a wrong submission csv",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 579947,
      "author_name": "arteam",
      "author_url": "",
      "post_date": "07/19/2019 12:31:23",
      "content": "<p>probably overfitting: You train a big model on a small dataset -&gt; big model just memorize all train images -&gt; bad on new images but really good at the train images</p>\n\n<p>Do you have used any validation data? Then you could see how well the models works on new images. \nIf not, try using a smaller model like ResNet-18 or 34 with some validation data.\nIf you are using validation data and the score is pretty good, you maybe generate a wrong submission csv</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "579884": "I have got 0.937 kappa score on train set, however on submission i got 0.014\nany idea why this is happening?\ni have used resent 152 model",
    "579947": "probably overfitting: You train a big model on a small dataset -&gt; big model just memorize all train images -&gt; bad on new images but really good at the train images\n\nDo you have used any validation data? Then you could see how well the models works on new images. \nIf not, try using a smaller model like ResNet-18 or 34 with some validation data.\nIf you are using validation data and the score is pretty good, you maybe generate a wrong submission csv"
  },
  "source": "meta"
}