{
  "id": 507969,
  "title": "[WITHOUT HACK] Best Private Score ",
  "url": "/competitions/home-credit-credit-risk-model-stability/discussion/507969",
  "author_name": "",
  "post_date": "2024-05-28T02:37:35.100633600Z",
  "votes": 4,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Okay. The end is funny and sad :) <br>\nHowever, can we as a community post our best non-hacked public and private scores, and the top ones if interested can share their notebooks, afterward? We would get to learn a lot.</p>\n<p>Mine is:<br>\nPublic LB: 0.602, Private LB: 0.526 (one of the selected final subs)<br>\n353 features, simple blend of LGB, CB and XGB</p>",
  "messages": [
    {
      "id": "2840193",
      "postDate": "05/28/2024 02:37:35",
      "content": "<p>Okay. The end is funny and sad :) <br>\nHowever, can we as a community post our best non-hacked public and private scores, and the top ones if interested can share their notebooks, afterward? We would get to learn a lot.</p>\n<p>Mine is:<br>\nPublic LB: 0.602, Private LB: 0.526 (one of the selected final subs)<br>\n353 features, simple blend of LGB, CB and XGB</p>",
      "rawMarkdown": "Okay. The end is funny and sad :) \nHowever, can we as a community post our best non-hacked public and private scores, and the top ones if interested can share their notebooks, afterward? We would get to learn a lot.\n\nMine is:\nPublic LB: 0.602, Private LB: 0.526 (one of the selected final subs)\n353 features, simple blend of LGB, CB and XGB",
      "votes": null
    },
    {
      "id": "2840198",
      "postDate": "05/28/2024 02:44:17",
      "content": "<p>The notebook I have been working on for 2 months - and perfecting it along the way - is the <a href=\"https://www.kaggle.com/code/andreasbis/crms-inference\" target=\"_blank\">CRMS: Inference</a>. The notebook scores 0.584/0.507 for public/private LB respectively. </p>\n<p>My approach was to imitate the best solutions from the AMEX competition two years ago. However, that did not go well: little did I know that cheating would be normalized as metric hacking…</p>",
      "rawMarkdown": "The notebook I have been working on for 2 months - and perfecting it along the way - is the [CRMS: Inference](https://www.kaggle.com/code/andreasbis/crms-inference). The notebook scores 0.584/0.507 for public/private LB respectively. \n\nMy approach was to imitate the best solutions from the AMEX competition two years ago. However, that did not go well: little did I know that cheating would be normalized as metric hacking...",
      "votes": null
    },
    {
      "id": "2840203",
      "postDate": "05/28/2024 02:51:45",
      "content": "<p>Public LB: 0.595, Private LB: 0.522</p>",
      "rawMarkdown": "Public LB: 0.595, Private LB: 0.522",
      "votes": null
    },
    {
      "id": "2840209",
      "postDate": "05/28/2024 02:58:27",
      "content": "<p>My best single and non-hacked model is XGB, 497 features (including hand-crafted features), 0.599 Public and 0.530 Private</p>",
      "rawMarkdown": "My best single and non-hacked model is XGB, 497 features (including hand-crafted features), 0.599 Public and 0.530 Private",
      "votes": null
    },
    {
      "id": "2840538",
      "postDate": "05/28/2024 06:13:47",
      "content": "<p>public:0.606, private:0.533</p>",
      "rawMarkdown": "public:0.606, private:0.533",
      "votes": null
    },
    {
      "id": "2841055",
      "postDate": "05/28/2024 11:30:30",
      "content": "<p>Public LB: 0.605, Private: 0.528<br>\n7 models ensemble, see <a href=\"https://www.kaggle.com/code/andreynesterov/home-credit-inference-final\" target=\"_blank\">Inference Notebook</a><br>\nP.S. BTW the best single model was CatBoost with a private score of 0.520 (unfortunately it wasn't included in the final ensemble).</p>",
      "rawMarkdown": "Public LB: 0.605, Private: 0.528\n7 models ensemble, see [Inference Notebook](https://www.kaggle.com/code/andreynesterov/home-credit-inference-final)\nP.S. BTW the best single model was CatBoost with a private score of 0.520 (unfortunately it wasn't included in the final ensemble).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2840198,
      "author_name": "andreasbis",
      "author_url": "",
      "post_date": "05/28/2024 02:44:17",
      "content": "<p>The notebook I have been working on for 2 months - and perfecting it along the way - is the <a href=\"https://www.kaggle.com/code/andreasbis/crms-inference\" target=\"_blank\">CRMS: Inference</a>. The notebook scores 0.584/0.507 for public/private LB respectively. </p>\n<p>My approach was to imitate the best solutions from the AMEX competition two years ago. However, that did not go well: little did I know that cheating would be normalized as metric hacking…</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2840203,
      "author_name": "vincentvvv",
      "author_url": "",
      "post_date": "05/28/2024 02:51:45",
      "content": "<p>Public LB: 0.595, Private LB: 0.522</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2840209,
      "author_name": "pntan17",
      "author_url": "",
      "post_date": "05/28/2024 02:58:27",
      "content": "<p>My best single and non-hacked model is XGB, 497 features (including hand-crafted features), 0.599 Public and 0.530 Private</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2840538,
      "author_name": "thunderthunder",
      "author_url": "",
      "post_date": "05/28/2024 06:13:47",
      "content": "<p>public:0.606, private:0.533</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 2841055,
      "author_name": "andreynesterov",
      "author_url": "",
      "post_date": "05/28/2024 11:30:30",
      "content": "<p>Public LB: 0.605, Private: 0.528<br>\n7 models ensemble, see <a href=\"https://www.kaggle.com/code/andreynesterov/home-credit-inference-final\" target=\"_blank\">Inference Notebook</a><br>\nP.S. BTW the best single model was CatBoost with a private score of 0.520 (unfortunately it wasn't included in the final ensemble).</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2840193": "Okay. The end is funny and sad :) \nHowever, can we as a community post our best non-hacked public and private scores, and the top ones if interested can share their notebooks, afterward? We would get to learn a lot.\n\nMine is:\nPublic LB: 0.602, Private LB: 0.526 (one of the selected final subs)\n353 features, simple blend of LGB, CB and XGB",
    "2840198": "The notebook I have been working on for 2 months - and perfecting it along the way - is the [CRMS: Inference](https://www.kaggle.com/code/andreasbis/crms-inference). The notebook scores 0.584/0.507 for public/private LB respectively. \n\nMy approach was to imitate the best solutions from the AMEX competition two years ago. However, that did not go well: little did I know that cheating would be normalized as metric hacking...",
    "2840203": "Public LB: 0.595, Private LB: 0.522",
    "2840209": "My best single and non-hacked model is XGB, 497 features (including hand-crafted features), 0.599 Public and 0.530 Private",
    "2840538": "public:0.606, private:0.533",
    "2841055": "Public LB: 0.605, Private: 0.528\n7 models ensemble, see [Inference Notebook](https://www.kaggle.com/code/andreynesterov/home-credit-inference-final)\nP.S. BTW the best single model was CatBoost with a private score of 0.520 (unfortunately it wasn't included in the final ensemble)."
  },
  "source": "meta"
}