{
  "id": 288419,
  "title": "43rd Place Solution",
  "content": "<h2>Thanks to</h2>\n<p>First of all, we'd like to thank kaggle and google research team. Also, thanks for everyone who participated in this competition. (especially,&nbsp;@kishalmandal and <a href=\"https://www.kaggle.com/tkm2261\" target=\"_blank\">@tkm2261</a>)</p>\n<h2>Solution</h2>\n<p>Model : deepset/xlm-roberta-large-squad2 (single)</p>\n<p>Data : </p>\n<ul>\n<li><code>Competition dataset + MLQA + XQUAD</code> (5 folds)</li>\n<li><code>Competition dataset + MLQA + XQUAD + squad_translated_tamil</code> (5 folds)</li>\n<li><code>Competition dataset + MLQA + XQUAD + squad_translated_tamil + Google translated SQuAD2.0 to Hindi and Tamil</code>  (2 folds)</li>\n</ul>\n<h2>Strategies</h2>\n<ol>\n<li><p>Soft voting with many folds was all we did. We assumed our 12 folds (Public LB: 0.778) will be better than 5 folds (Public LB: 0.792) on the Private LB since its folds seemed to be selected to perform well for Public LB only. Fortunately, It was exactly right. (12 folds Private LB: 0.740, 5 folds Private LB: 0.727)</p></li>\n<li><p>A couple of folds using additional data of Google translated SQuAD2.0 worked well for the ensemble (LB 0.004 higher), although the score of them was not that high. (Public LB 0.685, single fold submission) <br>\nWe guess this might contribute to our final submission's diversity of ingredients.</p></li>\n<li><p>To properly respond to the situation where 0.792s work well on Private LB, we added 0.792's 5 folds to our 12 folds and finalized the submission selection. (12 folds, 17 folds)  </p></li>\n<li><p>We tried Hard voting Ensemble, but its LB score was always lower than Soft voting. (0.001-0.005 lower)</p></li>\n</ol>",
  "comments": [
    {
      "id": 1585772,
      "author_name": "Kishal Mandal",
      "author_url": "",
      "post_date": "2021-11-17T14:14:18.773000",
      "content": "<p>Congratulations &lt;3. It seems like my model and notebooks were widely used. Thanks to all :) Hope I didnt let you guys down (except from the shakeup) :D</p>",
      "votes": 5,
      "replies": [
        {
          "id": 1586297,
          "author_name": "Dannykm",
          "author_url": "",
          "post_date": "2021-11-18T01:15:16.060000",
          "content": "<p>We learned a lot from you :)  Thank you for congratulations!</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1591242,
      "author_name": "Parv Yadav",
      "author_url": "",
      "post_date": "2021-11-22T05:45:09.553000",
      "content": "<p>Congratulations <a href=\"https://www.kaggle.com/dannykm\" target=\"_blank\">@dannykm</a> !✨</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1590702,
      "author_name": "MINGOOK KIM",
      "author_url": "",
      "post_date": "2021-11-21T15:10:46.957000",
      "content": "<p>Congratulations !</p>\n<p>Good Job!</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 1586695,
      "author_name": "MGC",
      "author_url": "",
      "post_date": "2021-11-18T08:17:03.213000",
      "content": "<p>Good work! It's helpful :)</p>",
      "votes": 1,
      "replies": [
        {
          "id": 1586697,
          "author_name": "Dannykm",
          "author_url": "",
          "post_date": "2021-11-18T08:18:51.257000",
          "content": "<p>Thanks bro :D</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "comment_count": 6,
  "messages": [
    {
      "id": 1585759,
      "postDate": "2021-11-17T13:59:40.007Z",
      "content": "<h2>Thanks to</h2>\n<p>First of all, we'd like to thank kaggle and google research team. Also, thanks for everyone who participated in this competition. (especially,&nbsp;@kishalmandal and <a href=\"https://www.kaggle.com/tkm2261\" target=\"_blank\">@tkm2261</a>)</p>\n<h2>Solution</h2>\n<p>Model : deepset/xlm-roberta-large-squad2 (single)</p>\n<p>Data : </p>\n<ul>\n<li><code>Competition dataset + MLQA + XQUAD</code> (5 folds)</li>\n<li><code>Competition dataset + MLQA + XQUAD + squad_translated_tamil</code> (5 folds)</li>\n<li><code>Competition dataset + MLQA + XQUAD + squad_translated_tamil + Google translated SQuAD2.0 to Hindi and Tamil</code>  (2 folds)</li>\n</ul>\n<h2>Strategies</h2>\n<ol>\n<li><p>Soft voting with many folds was all we did. We assumed our 12 folds (Public LB: 0.778) will be better than 5 folds (Public LB: 0.792) on the Private LB since its folds seemed to be selected to perform well for Public LB only. Fortunately, It was exactly right. (12 folds Private LB: 0.740, 5 folds Private LB: 0.727)</p></li>\n<li><p>A couple of folds using additional data of Google translated SQuAD2.0 worked well for the ensemble (LB 0.004 higher), although the score of them was not that high. (Public LB 0.685, single fold submission) <br>\nWe guess this might contribute to our final submission's diversity of ingredients.</p></li>\n<li><p>To properly respond to the situation where 0.792s work well on Private LB, we added 0.792's 5 folds to our 12 folds and finalized the submission selection. (12 folds, 17 folds)  </p></li>\n<li><p>We tried Hard voting Ensemble, but its LB score was always lower than Soft voting. (0.001-0.005 lower)</p></li>\n</ol>",
      "rawMarkdown": "## Thanks to\n\nFirst of all, we'd like to thank kaggle and google research team. Also, thanks for everyone who participated in this competition. (especially, @kishalmandal and @tkm2261)\n\n## Solution\n\nModel : deepset/xlm-roberta-large-squad2 (single)\n\nData : \n\n- `Competition dataset + MLQA + XQUAD` (5 folds)\n- `Competition dataset + MLQA + XQUAD + squad_translated_tamil` (5 folds)\n- `Competition dataset + MLQA + XQUAD + squad_translated_tamil + Google translated SQuAD2.0 to Hindi and Tamil`  (2 folds)\n\n## Strategies\n\n1. Soft voting with many folds was all we did. We assumed our 12 folds (Public LB: 0.778) will be better than 5 folds (Public LB: 0.792) on the Private LB since its folds seemed to be selected to perform well for Public LB only. Fortunately, It was exactly right. (12 folds Private LB: 0.740, 5 folds Private LB: 0.727)\n2. A couple of folds using additional data of Google translated SQuAD2.0 worked well for the ensemble (LB 0.004 higher), although the score of them was not that high. (Public LB 0.685, single fold submission) \nWe guess this might contribute to our final submission's diversity of ingredients.\n    \n3. To properly respond to the situation where 0.792s work well on Private LB, we added 0.792's 5 folds to our 12 folds and finalized the submission selection. (12 folds, 17 folds)  \n4. We tried Hard voting Ensemble, but its LB score was always lower than Soft voting. (0.001-0.005 lower)",
      "votes": 12
    },
    {
      "id": 1585772,
      "postDate": "2021-11-17T14:14:18.773Z",
      "content": "<p>Congratulations &lt;3. It seems like my model and notebooks were widely used. Thanks to all :) Hope I didnt let you guys down (except from the shakeup) :D</p>",
      "rawMarkdown": "Congratulations <3. It seems like my model and notebooks were widely used. Thanks to all :) Hope I didnt let you guys down (except from the shakeup) :D",
      "votes": 5,
      "replies": [
        {
          "id": 1586297,
          "postDate": "2021-11-18T01:15:16.060Z",
          "content": "<p>We learned a lot from you :)  Thank you for congratulations!</p>",
          "rawMarkdown": "We learned a lot from you :)  Thank you for congratulations!"
        }
      ]
    },
    {
      "id": 1591242,
      "postDate": "2021-11-22T05:45:09.553Z",
      "content": "<p>Congratulations <a href=\"https://www.kaggle.com/dannykm\" target=\"_blank\">@dannykm</a> !✨</p>",
      "rawMarkdown": "Congratulations @dannykm !✨",
      "votes": 1
    },
    {
      "id": 1590702,
      "postDate": "2021-11-21T15:10:46.957Z",
      "content": "<p>Congratulations !</p>\n<p>Good Job!</p>",
      "rawMarkdown": "Congratulations !\n\nGood Job!",
      "votes": 1
    },
    {
      "id": 1586695,
      "postDate": "2021-11-18T08:17:03.213Z",
      "content": "<p>Good work! It's helpful :)</p>",
      "rawMarkdown": "Good work! It's helpful :)",
      "votes": 1,
      "replies": [
        {
          "id": 1586697,
          "postDate": "2021-11-18T08:18:51.257Z",
          "content": "<p>Thanks bro :D</p>",
          "rawMarkdown": "Thanks bro :D"
        }
      ]
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-19T16:19:42Z",
  "competition_id": 30060,
  "competition_slug": "chaii-hindi-and-tamil-question-answering",
  "team_id": 7688710,
  "solution_url": "/competitions/chaii-hindi-and-tamil-question-answering/writeups/hlue-43rd-place-solution"
}