{
  "id": 175493,
  "title": "What a shake up !",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/175493",
  "author_name": "",
  "post_date": "2020-08-18T10:55:34.219831Z",
  "votes": 26,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello Everyone !</p>\n<p>I hope you are doing all good. The competition just ended few hours ago. I was in top 5% before the Private Leaderboard scoring. And it turns out that I finished around top 30%. I guess I mega overfit the LB data ! </p>\n<p>Tbh, i am glad, because now I have the evidence that a kaggle competition doesnt rely on many nonsense stacking levels.</p>\n<p>I was glad I took part at this competition and I learnt a lot, which is actually the most important thing !</p>",
  "messages": [
    {
      "id": "975553",
      "postDate": "08/18/2020 10:55:34",
      "content": "<p>Hello Everyone !</p>\n<p>I hope you are doing all good. The competition just ended few hours ago. I was in top 5% before the Private Leaderboard scoring. And it turns out that I finished around top 30%. I guess I mega overfit the LB data ! </p>\n<p>Tbh, i am glad, because now I have the evidence that a kaggle competition doesnt rely on many nonsense stacking levels.</p>\n<p>I was glad I took part at this competition and I learnt a lot, which is actually the most important thing !</p>",
      "rawMarkdown": "Hello Everyone !\n\nI hope you are doing all good. The competition just ended few hours ago. I was in top 5% before the Private Leaderboard scoring. And it turns out that I finished around top 30%. I guess I mega overfit the LB data ! \n\nTbh, i am glad, because now I have the evidence that a kaggle competition doesnt rely on many nonsense stacking levels.\n\nI was glad I took part at this competition and I learnt a lot, which is actually the most important thing !",
      "votes": null
    },
    {
      "id": "976736",
      "postDate": "08/19/2020 04:18:21",
      "content": "<p>This is normal, I think most of the 95%+ accuracy are just overfitting their models… It happened to many of my competitions so my team simply stopped at 96%, we knew that we did what we could. Learning is the gold medal here… as for the prices in a competition, is still a mystery for me… if data changes drastically, how good this layers would be holding to retrain on the new data? Does the new data need additional scripting to modify the input to match the current data ?</p>",
      "rawMarkdown": "This is normal, I think most of the 95%+ accuracy are just overfitting their models... It happened to many of my competitions so my team simply stopped at 96%, we knew that we did what we could. Learning is the gold medal here... as for the prices in a competition, is still a mystery for me... if data changes drastically, how good this layers would be holding to retrain on the new data? Does the new data need additional scripting to modify the input to match the current data ?",
      "votes": null
    },
    {
      "id": "979114",
      "postDate": "08/20/2020 16:16:30",
      "content": "<p>You made a very nice point</p>",
      "rawMarkdown": "You made a very nice point",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 976736,
      "author_name": "alincijov",
      "author_url": "",
      "post_date": "08/19/2020 04:18:21",
      "content": "<p>This is normal, I think most of the 95%+ accuracy are just overfitting their models… It happened to many of my competitions so my team simply stopped at 96%, we knew that we did what we could. Learning is the gold medal here… as for the prices in a competition, is still a mystery for me… if data changes drastically, how good this layers would be holding to retrain on the new data? Does the new data need additional scripting to modify the input to match the current data ?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 979114,
      "author_name": "subham07",
      "author_url": "",
      "post_date": "08/20/2020 16:16:30",
      "content": "<p>You made a very nice point</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "975553": "Hello Everyone !\n\nI hope you are doing all good. The competition just ended few hours ago. I was in top 5% before the Private Leaderboard scoring. And it turns out that I finished around top 30%. I guess I mega overfit the LB data ! \n\nTbh, i am glad, because now I have the evidence that a kaggle competition doesnt rely on many nonsense stacking levels.\n\nI was glad I took part at this competition and I learnt a lot, which is actually the most important thing !",
    "976736": "This is normal, I think most of the 95%+ accuracy are just overfitting their models... It happened to many of my competitions so my team simply stopped at 96%, we knew that we did what we could. Learning is the gold medal here... as for the prices in a competition, is still a mystery for me... if data changes drastically, how good this layers would be holding to retrain on the new data? Does the new data need additional scripting to modify the input to match the current data ?",
    "979114": "You made a very nice point"
  },
  "source": "meta"
}