{
  "id": 340698,
  "title": "There is a silver solution hidden in public notebooks!",
  "url": "/competitions/smartphone-decimeter-2022/discussion/340698",
  "author_name": "ForcewithMe",
  "post_date": "2022-07-30T14:41:52.555000",
  "votes": 9,
  "comment_count": 0,
  "views": 0,
  "content": "<p>First, thanks <a href=\"https://www.kaggle.com/timeverett\" target=\"_blank\">@timeverett</a> for your amazing rtk notebook. </p>\n<p><a href=\"https://www.kaggle.com/timeverett\" target=\"_blank\">@timeverett</a> published two rtk notebooks, which used <a href=\"https://www.kaggle.com/code/timeverett/getting-started-with-rtklib\" target=\"_blank\">C</a> and <a href=\"https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py\" target=\"_blank\">python</a> to implement functions respectively.</p>\n<ul>\n<li><p>The latest version of C can score 2.679 on LB, 2.189 on PB(Bronze medal) without any modification.</p></li>\n<li><p>The python notebook hasn't been updated for a long time, it can score 3.385 on LB, <strong>but 2.107 on PB</strong>! It can bring you a silver medal without any modification.</p></li>\n</ul>\n<p>If you calculate the scores of each kind of phone on training set, you can find that it performs really badly on GooglePixel4, and GooglePixel4XL. There are no any 4XLs in test set, 3 Pixel4s in LB, but no any 4XL and Pixel4 in PB. You may be able to guess what I want to say now, even if you don't know anything about this competition. It only scores 3.385 on LB because of Pixel4. If you replace the Pixel4 of the rtk baseline(python version) with any other public baseline scored ~3.0, it can score ~2.6 on LB. </p>\n<p>In fact, we have modified taro's public baseline and improved it from 3.027 to 2.66. But still, the rtk baseline(python, scored 3.385 on LB) has a better score for Samsung and GooglePixel5 on the training set than the 2.66 baseline. Samsung and GooglePixel5 take 7 out of 8 paths from PB!</p>",
  "messages": [
    {
      "id": 1877334,
      "postDate": "2022-07-30T14:41:52.557Z",
      "content": "<p>First, thanks <a href=\"https://www.kaggle.com/timeverett\" target=\"_blank\">@timeverett</a> for your amazing rtk notebook. </p>\n<p><a href=\"https://www.kaggle.com/timeverett\" target=\"_blank\">@timeverett</a> published two rtk notebooks, which used <a href=\"https://www.kaggle.com/code/timeverett/getting-started-with-rtklib\" target=\"_blank\">C</a> and <a href=\"https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py\" target=\"_blank\">python</a> to implement functions respectively.</p>\n<ul>\n<li><p>The latest version of C can score 2.679 on LB, 2.189 on PB(Bronze medal) without any modification.</p></li>\n<li><p>The python notebook hasn't been updated for a long time, it can score 3.385 on LB, <strong>but 2.107 on PB</strong>! It can bring you a silver medal without any modification.</p></li>\n</ul>\n<p>If you calculate the scores of each kind of phone on training set, you can find that it performs really badly on GooglePixel4, and GooglePixel4XL. There are no any 4XLs in test set, 3 Pixel4s in LB, but no any 4XL and Pixel4 in PB. You may be able to guess what I want to say now, even if you don't know anything about this competition. It only scores 3.385 on LB because of Pixel4. If you replace the Pixel4 of the rtk baseline(python version) with any other public baseline scored ~3.0, it can score ~2.6 on LB. </p>\n<p>In fact, we have modified taro's public baseline and improved it from 3.027 to 2.66. But still, the rtk baseline(python, scored 3.385 on LB) has a better score for Samsung and GooglePixel5 on the training set than the 2.66 baseline. Samsung and GooglePixel5 take 7 out of 8 paths from PB!</p>",
      "rawMarkdown": "First, thanks @timeverett for your amazing rtk notebook. \n\n@timeverett published two rtk notebooks, which used [C](https://www.kaggle.com/code/timeverett/getting-started-with-rtklib) and [python](https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py) to implement functions respectively.\n\n- The latest version of C can score 2.679 on LB, 2.189 on PB(Bronze medal) without any modification.\n\n- The python notebook hasn't been updated for a long time, it can score 3.385 on LB, **but 2.107 on PB**! It can bring you a silver medal without any modification.\n\nIf you calculate the scores of each kind of phone on training set, you can find that it performs really badly on GooglePixel4, and GooglePixel4XL. There are no any 4XLs in test set, 3 Pixel4s in LB, but no any 4XL and Pixel4 in PB. You may be able to guess what I want to say now, even if you don't know anything about this competition. It only scores 3.385 on LB because of Pixel4. If you replace the Pixel4 of the rtk baseline(python version) with any other public baseline scored ~3.0, it can score ~2.6 on LB. \n\nIn fact, we have modified taro's public baseline and improved it from 3.027 to 2.66. But still, the rtk baseline(python, scored 3.385 on LB) has a better score for Samsung and GooglePixel5 on the training set than the 2.66 baseline. Samsung and GooglePixel5 take 7 out of 8 paths from PB!",
      "votes": 9
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1877334": "First, thanks @timeverett for your amazing rtk notebook. \n\n@timeverett published two rtk notebooks, which used [C](https://www.kaggle.com/code/timeverett/getting-started-with-rtklib) and [python](https://www.kaggle.com/code/timeverett/getting-started-with-rtklib-py) to implement functions respectively.\n\n- The latest version of C can score 2.679 on LB, 2.189 on PB(Bronze medal) without any modification.\n\n- The python notebook hasn't been updated for a long time, it can score 3.385 on LB, **but 2.107 on PB**! It can bring you a silver medal without any modification.\n\nIf you calculate the scores of each kind of phone on training set, you can find that it performs really badly on GooglePixel4, and GooglePixel4XL. There are no any 4XLs in test set, 3 Pixel4s in LB, but no any 4XL and Pixel4 in PB. You may be able to guess what I want to say now, even if you don't know anything about this competition. It only scores 3.385 on LB because of Pixel4. If you replace the Pixel4 of the rtk baseline(python version) with any other public baseline scored ~3.0, it can score ~2.6 on LB. \n\nIn fact, we have modified taro's public baseline and improved it from 3.027 to 2.66. But still, the rtk baseline(python, scored 3.385 on LB) has a better score for Samsung and GooglePixel5 on the training set than the 2.66 baseline. Samsung and GooglePixel5 take 7 out of 8 paths from PB!"
  }
}