{
  "id": 518786,
  "title": "Good luck to everyone for the shake/shuffle in private LB",
  "url": "/competitions/leash-BELKA/discussion/518786",
  "author_name": "Yifan Wu",
  "post_date": "2024-07-08T08:41:05.908000",
  "votes": 6,
  "comment_count": 14,
  "views": 0,
  "content": "<p>By the way, what's the beset score a single mode can get?<br>\nMy best single model (using fingerprint/atom-level/SMILES-level features) is 0.458 in public LB. Does anyone have a better one? </p>",
  "messages": [
    {
      "id": 2911353,
      "postDate": "2024-07-08T08:41:05.910Z",
      "content": "<p>By the way, what's the beset score a single mode can get?<br>\nMy best single model (using fingerprint/atom-level/SMILES-level features) is 0.458 in public LB. Does anyone have a better one? </p>",
      "rawMarkdown": "By the way, what's the beset score a single mode can get?\nMy best single model (using fingerprint/atom-level/SMILES-level features) is 0.458 in public LB. Does anyone have a better one? ",
      "votes": 6
    },
    {
      "id": 2912504,
      "postDate": "2024-07-09T00:15:45.540Z",
      "content": "<p>good luck! what we can do is just praying.</p>",
      "rawMarkdown": "good luck! what we can do is just praying.",
      "votes": 1
    },
    {
      "id": 2912506,
      "postDate": "2024-07-09T00:17:02.737Z",
      "content": "<p>crazy shake!</p>",
      "rawMarkdown": "crazy shake!",
      "votes": 2,
      "replies": [
        {
          "id": 2912507,
          "postDate": "2024-07-09T00:19:49.987Z",
          "content": "<p>Congrats!!!  At least, you still get the silver.  </p>",
          "rawMarkdown": "Congrats!!!  At least, you still get the silver.  "
        }
      ]
    },
    {
      "id": 2911388,
      "postDate": "2024-07-08T09:06:01.773Z",
      "content": "<p>Single model: 0.466 fold0 and 0.477 (Train with all data).   Other folds can only get 0.440 at the best. I don't have other good things to average with.  I should have merge with others to form a team.</p>",
      "rawMarkdown": "Single model: 0.466 fold0 and 0.477 (Train with all data).   Other folds can only get 0.440 at the best. I don't have other good things to average with.  I should have merge with others to form a team.",
      "votes": 2,
      "replies": [
        {
          "id": 2911672,
          "postDate": "2024-07-08T13:21:11.750Z",
          "content": "<p>Wow. Can I ask what features do you use? 0.477 for a single model is a pretty high score. If you try to ensemble more diffearent models, I think you'll easily reach to 0.490!</p>",
          "rawMarkdown": "Wow. Can I ask what features do you use? 0.477 for a single model is a pretty high score. If you try to ensemble more diffearent models, I think you'll easily reach to 0.490!",
          "replies": [
            {
              "id": 2911712,
              "postDate": "2024-07-08T13:54:49.853Z",
              "content": "<p>I used GNN. </p>",
              "rawMarkdown": "I used GNN. ",
              "votes": 1
            },
            {
              "id": 2911909,
              "postDate": "2024-07-08T15:54:36.783Z",
              "content": "<p>Great, I also tried GNN. The offline score is 0.638 (one of the 15fold) but 0.398 LB. <br>\nLooking forward to learn your solution after the competition if you'd like to share it!</p>",
              "rawMarkdown": "Great, I also tried GNN. The offline score is 0.638 (one of the 15fold) but 0.398 LB. \nLooking forward to learn your solution after the competition if you'd like to share it!",
              "votes": 1
            },
            {
              "id": 2912310,
              "postDate": "2024-07-08T19:36:26.003Z",
              "content": "<p>I think your GNN might not be too far away from mine.  Some of my folds are 0.37-0.4.  I think 0.466 might be overfitting public LB, not really too much different from 0.440. Also, I used 7 edge features.  I just saw a public notebook used 11 edge features.  Wonder more edge features would be good. <a href=\"https://www.kaggle.com/code/joshualin24/graph-neural-network-baseline/comments?scriptVersionId=180166849\" target=\"_blank\">https://www.kaggle.com/code/joshualin24/graph-neural-network-baseline/comments?scriptVersionId=180166849</a></p>",
              "rawMarkdown": "I think your GNN might not be too far away from mine.  Some of my folds are 0.37-0.4.  I think 0.466 might be overfitting public LB, not really too much different from 0.440. Also, I used 7 edge features.  I just saw a public notebook used 11 edge features.  Wonder more edge features would be good. https://www.kaggle.com/code/joshualin24/graph-neural-network-baseline/comments?scriptVersionId=180166849"
            },
            {
              "id": 2912494,
              "postDate": "2024-07-09T00:08:06.543Z",
              "content": "<p>I dropped more than 1000 places.  I guess my method overfits the PLB.</p>",
              "rawMarkdown": "I dropped more than 1000 places.  I guess my method overfits the PLB."
            },
            {
              "id": 2912510,
              "postDate": "2024-07-09T00:28:02.097Z",
              "content": "<p>Thank you for your sharing. I don't think this is caused by so-called overfitting. Overfitting will not let a more than 1000 ranks of public LB be the top rank in private LB. I have participated similar drug/protein competition, and see the shuffle in private LB every time. I'm so disappointed about these kinds of protein/molecule competition. </p>",
              "rawMarkdown": "Thank you for your sharing. I don't think this is caused by so-called overfitting. Overfitting will not let a more than 1000 ranks of public LB be the top rank in private LB. I have participated similar drug/protein competition, and see the shuffle in private LB every time. I'm so disappointed about these kinds of protein/molecule competition. ",
              "votes": 1
            }
          ]
        }
      ]
    },
    {
      "id": 2912513,
      "postDate": "2024-07-09T00:32:58.360Z",
      "rawMarkdown": "",
      "votes": 2,
      "isDeleted": true,
      "replies": [
        {
          "id": 2912520,
          "postDate": "2024-07-09T00:37:31.410Z",
          "content": "<p>Mine was 0.477 for the public. Very bad for the private 0.212. </p>",
          "rawMarkdown": "Mine was 0.477 for the public. Very bad for the private 0.212. "
        },
        {
          "id": 2912521,
          "postDate": "2024-07-09T00:38:13.187Z",
          "content": "<p>cool! how about your private LB result? is it the 0.474 GNN?</p>",
          "rawMarkdown": "cool! how about your private LB result? is it the 0.474 GNN?"
        }
      ]
    },
    {
      "id": 2912501,
      "postDate": "2024-07-09T00:13:31.477Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 2912504,
      "author_name": "kyu999",
      "author_url": "",
      "post_date": "2024-07-09T00:15:45.540000",
      "content": "<p>good luck! what we can do is just praying.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2912506,
      "author_name": "Yifan Wu",
      "author_url": "",
      "post_date": "2024-07-09T00:17:02.737000",
      "content": "<p>crazy shake!</p>",
      "votes": 2,
      "replies": [
        {
          "id": 2912507,
          "author_name": "joejeo1",
          "author_url": "",
          "post_date": "2024-07-09T00:19:49.987000",
          "content": "<p>Congrats!!!  At least, you still get the silver.  </p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2911388,
      "author_name": "joejeo1",
      "author_url": "",
      "post_date": "2024-07-08T09:06:01.773000",
      "content": "<p>Single model: 0.466 fold0 and 0.477 (Train with all data).   Other folds can only get 0.440 at the best. I don't have other good things to average with.  I should have merge with others to form a team.</p>",
      "votes": 2,
      "replies": [
        {
          "id": 2911672,
          "author_name": "Yifan Wu",
          "author_url": "",
          "post_date": "2024-07-08T13:21:11.750000",
          "content": "<p>Wow. Can I ask what features do you use? 0.477 for a single model is a pretty high score. If you try to ensemble more diffearent models, I think you'll easily reach to 0.490!</p>",
          "votes": 0,
          "replies": [
            {
              "id": 2911712,
              "author_name": "joejeo1",
              "author_url": "",
              "post_date": "2024-07-08T13:54:49.853000",
              "content": "<p>I used GNN. </p>",
              "votes": 1,
              "replies": []
            },
            {
              "id": 2911909,
              "author_name": "Yifan Wu",
              "author_url": "",
              "post_date": "2024-07-08T15:54:36.783000",
              "content": "<p>Great, I also tried GNN. The offline score is 0.638 (one of the 15fold) but 0.398 LB. <br>\nLooking forward to learn your solution after the competition if you'd like to share it!</p>",
              "votes": 1,
              "replies": []
            },
            {
              "id": 2912310,
              "author_name": "joejeo1",
              "author_url": "",
              "post_date": "2024-07-08T19:36:26.003000",
              "content": "<p>I think your GNN might not be too far away from mine.  Some of my folds are 0.37-0.4.  I think 0.466 might be overfitting public LB, not really too much different from 0.440. Also, I used 7 edge features.  I just saw a public notebook used 11 edge features.  Wonder more edge features would be good. <a href=\"https://www.kaggle.com/code/joshualin24/graph-neural-network-baseline/comments?scriptVersionId=180166849\" target=\"_blank\">https://www.kaggle.com/code/joshualin24/graph-neural-network-baseline/comments?scriptVersionId=180166849</a></p>",
              "votes": 0,
              "replies": []
            },
            {
              "id": 2912494,
              "author_name": "joejeo1",
              "author_url": "",
              "post_date": "2024-07-09T00:08:06.543000",
              "content": "<p>I dropped more than 1000 places.  I guess my method overfits the PLB.</p>",
              "votes": 0,
              "replies": []
            },
            {
              "id": 2912510,
              "author_name": "Yifan Wu",
              "author_url": "",
              "post_date": "2024-07-09T00:28:02.097000",
              "content": "<p>Thank you for your sharing. I don't think this is caused by so-called overfitting. Overfitting will not let a more than 1000 ranks of public LB be the top rank in private LB. I have participated similar drug/protein competition, and see the shuffle in private LB every time. I'm so disappointed about these kinds of protein/molecule competition. </p>",
              "votes": 1,
              "replies": []
            }
          ]
        }
      ]
    },
    {
      "id": 2912513,
      "author_name": "",
      "author_url": "",
      "post_date": "2024-07-09T00:32:58.360000",
      "content": "",
      "votes": 2,
      "replies": [
        {
          "id": 2912520,
          "author_name": "joejeo1",
          "author_url": "",
          "post_date": "2024-07-09T00:37:31.410000",
          "content": "<p>Mine was 0.477 for the public. Very bad for the private 0.212. </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 2912521,
          "author_name": "Yifan Wu",
          "author_url": "",
          "post_date": "2024-07-09T00:38:13.187000",
          "content": "<p>cool! how about your private LB result? is it the 0.474 GNN?</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2912501,
      "author_name": "",
      "author_url": "",
      "post_date": "2024-07-09T00:13:31.477000",
      "content": "",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2911353": "By the way, what's the beset score a single mode can get?\nMy best single model (using fingerprint/atom-level/SMILES-level features) is 0.458 in public LB. Does anyone have a better one? ",
    "2912504": "good luck! what we can do is just praying.",
    "2912506": "crazy shake!",
    "2911388": "Single model: 0.466 fold0 and 0.477 (Train with all data).   Other folds can only get 0.440 at the best. I don't have other good things to average with.  I should have merge with others to form a team.",
    "2912513": "",
    "2912501": ""
  }
}