{
  "id": 347383,
  "title": "72th short solution (2 stage model)",
  "content": "<p>Congrats and thanks all kaggler sharing, This is my first solo silver medal. I learn a lot in this competition.</p>\n<p>Brief solution as follows</p>\n<ol>\n<li>My model input like <a href=\"https://www.kaggle.com/code/nbroad/token-classification-approach-fpe\" target=\"_blank\">https://www.kaggle.com/code/nbroad/token-classification-approach-fpe</a>, thanks <a href=\"https://www.kaggle.com/nbroad\" target=\"_blank\">@nbroad</a> </li>\n<li>use xgb and lgb model and add some handcraft features, it can imporve your score about 0.01</li>\n</ol>\n<p>final model :<br>\nDeberta-V3-Large + Deberta-Large (1 stage model) <br>\nPrivate : 0.597 Public 0.591</p>\n<p>Deberta-V3-Large + Deberta-Large (2 stage model lgb+xgb) <br>\nPrivate : 0.586 Public 0.586 (But I don't select this in my final :( ) </p>\n<p>Deberta-V3-Large + Deberta-Large + Roberta-Large (2 stage model  only lgb) <br>\nPrivate : 0.588 Public 0.585  </p>\n<p>HandCraft additional Feature as belows :<br>\n(reference Automated essay scoring paper)</p>\n<p>feature_col = [</p>\n<pre><code>    'adequate', 'effective', 'ineffective',  # deberta predict\n\n    'essay_len', 'discourse_type_amt_nunique','discourse_type_cnt',\n   'discourse_type_shift', 'text_len',\n    # textstat \n   'syllable_count',\n   'flesch_reading_ease', 'flesch_kincaid_grade', 'fog_scale', 'smog',\n   'automated_readability', 'coleman_liau', 'linsear_write',\n   'dale_chall_readability', 'text_standard', 'total_words', 'stop_prop',\n   'unique_words', 'spelling_mistake_count',\n\n   'difficult_words', 'crawford'\n    ',', '.', 'VB', 'JJR', 'WP', ### pos tag features\n   'PRP$', 'VBN', 'VBG', 'IN', 'CC', 'JJS', 'PRP', 'MD', 'WRB', 'RB',\n   'VBD', 'RBR', 'VBZ', 'NNP', 'POS', 'WDT', 'DT', 'CD', 'NN', 'TO', 'JJ',\n   'VBP', 'RP', 'NNS',]\n</code></pre>",
  "comments": [
    {
      "id": 1911307,
      "author_name": "Nicholas Broad",
      "author_url": "",
      "post_date": "2022-08-24T02:41:50.503000",
      "content": "<p>Congrats on your first solo silver! 🎉</p>\n<p>I'm happy to hear my notebook was helpful 🤗 </p>",
      "votes": 3,
      "replies": [
        {
          "id": 1911326,
          "author_name": "arvis",
          "author_url": "",
          "post_date": "2022-08-24T03:03:25.550000",
          "content": "<p>thanks a lot 😃<br>\nyour notebook are informative, and learn lots of this</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1911263,
      "author_name": "AK",
      "author_url": "",
      "post_date": "2022-08-24T01:58:50.440000",
      "content": "<p>you used readiblity features from commonlit? how much is the boost you got from adding those features </p>",
      "votes": 2,
      "replies": [
        {
          "id": 1911284,
          "author_name": "arvis",
          "author_url": "",
          "post_date": "2022-08-24T02:17:17.390000",
          "content": "<p>I dont try  only use deberta predict prob in 2 stage model, but in my local cv</p>\n<p>if only use deberta model to 2 stage model (no additional features) CV : 0.601~0.602</p>\n<p>if add addtional features CV : 0.595~0.596</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 1911913,
          "author_name": "AK",
          "author_url": "",
          "post_date": "2022-08-24T11:41:15.353000",
          "content": "<p>Thanks nice finish :)</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 1912090,
      "author_name": "Patrick Chan",
      "author_url": "",
      "post_date": "2022-08-24T13:52:09.270000",
      "content": "<p>Thanks for sharing.  Can you help clarify where is this link of information you mentioned: \"HandCraft additional Feature as belows :<br>\n(reference Automated essay scoring paper)\"? </p>",
      "votes": 0,
      "replies": [
        {
          "id": 1912101,
          "author_name": "arvis",
          "author_url": "",
          "post_date": "2022-08-24T14:05:31.567000",
          "content": "<p>sure, I picked out some automate essay scoring paper, you can read<br>\n[<a href=\"https://arxiv.org/pdf/1901.07744.pdf\" target=\"_blank\">https://arxiv.org/pdf/1901.07744.pdf</a>]<br>\n[<a href=\"https://arxiv.org/pdf/2008.01441.pdf\" target=\"_blank\">https://arxiv.org/pdf/2008.01441.pdf</a>]</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "comment_count": 7,
  "messages": [
    {
      "id": 1911225,
      "postDate": "2022-08-24T01:26:30.957Z",
      "content": "<p>Congrats and thanks all kaggler sharing, This is my first solo silver medal. I learn a lot in this competition.</p>\n<p>Brief solution as follows</p>\n<ol>\n<li>My model input like <a href=\"https://www.kaggle.com/code/nbroad/token-classification-approach-fpe\" target=\"_blank\">https://www.kaggle.com/code/nbroad/token-classification-approach-fpe</a>, thanks <a href=\"https://www.kaggle.com/nbroad\" target=\"_blank\">@nbroad</a> </li>\n<li>use xgb and lgb model and add some handcraft features, it can imporve your score about 0.01</li>\n</ol>\n<p>final model :<br>\nDeberta-V3-Large + Deberta-Large (1 stage model) <br>\nPrivate : 0.597 Public 0.591</p>\n<p>Deberta-V3-Large + Deberta-Large (2 stage model lgb+xgb) <br>\nPrivate : 0.586 Public 0.586 (But I don't select this in my final :( ) </p>\n<p>Deberta-V3-Large + Deberta-Large + Roberta-Large (2 stage model  only lgb) <br>\nPrivate : 0.588 Public 0.585  </p>\n<p>HandCraft additional Feature as belows :<br>\n(reference Automated essay scoring paper)</p>\n<p>feature_col = [</p>\n<pre><code>    'adequate', 'effective', 'ineffective',  # deberta predict\n\n    'essay_len', 'discourse_type_amt_nunique','discourse_type_cnt',\n   'discourse_type_shift', 'text_len',\n    # textstat \n   'syllable_count',\n   'flesch_reading_ease', 'flesch_kincaid_grade', 'fog_scale', 'smog',\n   'automated_readability', 'coleman_liau', 'linsear_write',\n   'dale_chall_readability', 'text_standard', 'total_words', 'stop_prop',\n   'unique_words', 'spelling_mistake_count',\n\n   'difficult_words', 'crawford'\n    ',', '.', 'VB', 'JJR', 'WP', ### pos tag features\n   'PRP$', 'VBN', 'VBG', 'IN', 'CC', 'JJS', 'PRP', 'MD', 'WRB', 'RB',\n   'VBD', 'RBR', 'VBZ', 'NNP', 'POS', 'WDT', 'DT', 'CD', 'NN', 'TO', 'JJ',\n   'VBP', 'RP', 'NNS',]\n</code></pre>",
      "rawMarkdown": "Congrats and thanks all kaggler sharing, This is my first solo silver medal. I learn a lot in this competition.\n\nBrief solution as follows\n1. My model input like https://www.kaggle.com/code/nbroad/token-classification-approach-fpe, thanks @nbroad \n2. use xgb and lgb model and add some handcraft features, it can imporve your score about 0.01\n\nfinal model :\nDeberta-V3-Large + Deberta-Large (1 stage model) \nPrivate : 0.597 Public 0.591\n\nDeberta-V3-Large + Deberta-Large (2 stage model lgb+xgb) \nPrivate : 0.586 Public 0.586 (But I don't select this in my final :( ) \n\nDeberta-V3-Large + Deberta-Large + Roberta-Large (2 stage model  only lgb) \nPrivate : 0.588 Public 0.585  \n\n\nHandCraft additional Feature as belows :\n(reference Automated essay scoring paper)\n\nfeature_col = [\n\n        'adequate', 'effective', 'ineffective',  # deberta predict\n\n        'essay_len', 'discourse_type_amt_nunique','discourse_type_cnt',\n       'discourse_type_shift', 'text_len',\n        # textstat \n       'syllable_count',\n       'flesch_reading_ease', 'flesch_kincaid_grade', 'fog_scale', 'smog',\n       'automated_readability', 'coleman_liau', 'linsear_write',\n       'dale_chall_readability', 'text_standard', 'total_words', 'stop_prop',\n       'unique_words', 'spelling_mistake_count',\n       \n       'difficult_words', 'crawford'\n        ',', '.', 'VB', 'JJR', 'WP', ### pos tag features\n       'PRP$', 'VBN', 'VBG', 'IN', 'CC', 'JJS', 'PRP', 'MD', 'WRB', 'RB',\n       'VBD', 'RBR', 'VBZ', 'NNP', 'POS', 'WDT', 'DT', 'CD', 'NN', 'TO', 'JJ',\n       'VBP', 'RP', 'NNS',]\n\n\n \n",
      "votes": 15
    },
    {
      "id": 1911307,
      "postDate": "2022-08-24T02:41:50.503Z",
      "content": "<p>Congrats on your first solo silver! 🎉</p>\n<p>I'm happy to hear my notebook was helpful 🤗 </p>",
      "rawMarkdown": "Congrats on your first solo silver! 🎉\n\nI'm happy to hear my notebook was helpful 🤗 ",
      "votes": 3,
      "replies": [
        {
          "id": 1911326,
          "postDate": "2022-08-24T03:03:25.550Z",
          "content": "<p>thanks a lot 😃<br>\nyour notebook are informative, and learn lots of this</p>",
          "rawMarkdown": "thanks a lot 😃\nyour notebook are informative, and learn lots of this"
        }
      ]
    },
    {
      "id": 1911263,
      "postDate": "2022-08-24T01:58:50.440Z",
      "content": "<p>you used readiblity features from commonlit? how much is the boost you got from adding those features </p>",
      "rawMarkdown": "you used readiblity features from commonlit? how much is the boost you got from adding those features ",
      "votes": 2,
      "replies": [
        {
          "id": 1911284,
          "postDate": "2022-08-24T02:17:17.390Z",
          "content": "<p>I dont try  only use deberta predict prob in 2 stage model, but in my local cv</p>\n<p>if only use deberta model to 2 stage model (no additional features) CV : 0.601~0.602</p>\n<p>if add addtional features CV : 0.595~0.596</p>",
          "rawMarkdown": "I dont try  only use deberta predict prob in 2 stage model, but in my local cv\n\nif only use deberta model to 2 stage model (no additional features) CV : 0.601~0.602\n\nif add addtional features CV : 0.595~0.596",
          "votes": 1
        },
        {
          "id": 1911913,
          "postDate": "2022-08-24T11:41:15.353Z",
          "content": "<p>Thanks nice finish :)</p>",
          "rawMarkdown": "Thanks nice finish :)\n"
        }
      ]
    },
    {
      "id": 1912090,
      "postDate": "2022-08-24T13:52:09.270Z",
      "content": "<p>Thanks for sharing.  Can you help clarify where is this link of information you mentioned: \"HandCraft additional Feature as belows :<br>\n(reference Automated essay scoring paper)\"? </p>",
      "rawMarkdown": "Thanks for sharing.  Can you help clarify where is this link of information you mentioned: \"HandCraft additional Feature as belows :\n(reference Automated essay scoring paper)\"? ",
      "replies": [
        {
          "id": 1912101,
          "postDate": "2022-08-24T14:05:31.567Z",
          "content": "<p>sure, I picked out some automate essay scoring paper, you can read<br>\n[<a href=\"https://arxiv.org/pdf/1901.07744.pdf\" target=\"_blank\">https://arxiv.org/pdf/1901.07744.pdf</a>]<br>\n[<a href=\"https://arxiv.org/pdf/2008.01441.pdf\" target=\"_blank\">https://arxiv.org/pdf/2008.01441.pdf</a>]</p>",
          "rawMarkdown": "sure, I picked out some automate essay scoring paper, you can read\n[https://arxiv.org/pdf/1901.07744.pdf]\n[https://arxiv.org/pdf/2008.01441.pdf]",
          "votes": 1
        }
      ]
    }
  ],
  "complete": true,
  "source": "live",
  "fetched_at": "2026-10-19T17:54:50Z",
  "competition_id": 35308,
  "competition_slug": "feedback-prize-effectiveness",
  "team_id": 8752275,
  "solution_url": "/competitions/feedback-prize-effectiveness/writeups/grrr-72th-short-solution-2-stage-model"
}