{
  "id": 211321,
  "title": "Attempts to train multiclass against actual user_answer[0..3]?",
  "url": "/competitions/riiid-test-answer-prediction/discussion/211321",
  "author_name": "",
  "post_date": "2021-01-14T16:18:04.939373Z",
  "votes": 3,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi, I am curious to hear about any attempts to set up and train auxillary models against actually given user answers as part of a greater solution?</p>\n<p>I did not find time to investigate this myself. It sounds 'less wrong' to opt for 'Stockholm' as the capital of Norway rather than 'Kuala Lumpur', and I am wondering if there might as such be signal in the actual value of 'user_answer'. Thanks!</p>",
  "messages": [
    {
      "id": "1153123",
      "postDate": "01/14/2021 16:18:04",
      "content": "<p>Hi, I am curious to hear about any attempts to set up and train auxillary models against actually given user answers as part of a greater solution?</p>\n<p>I did not find time to investigate this myself. It sounds 'less wrong' to opt for 'Stockholm' as the capital of Norway rather than 'Kuala Lumpur', and I am wondering if there might as such be signal in the actual value of 'user_answer'. Thanks!</p>",
      "rawMarkdown": "Hi, I am curious to hear about any attempts to set up and train auxillary models against actually given user answers as part of a greater solution?\n\nI did not find time to investigate this myself. It sounds 'less wrong' to opt for 'Stockholm' as the capital of Norway rather than 'Kuala Lumpur', and I am wondering if there might as such be signal in the actual value of 'user_answer'. Thanks!",
      "votes": null
    },
    {
      "id": "1153226",
      "postDate": "01/14/2021 17:40:53",
      "content": "<p>I did it, it helps my model training less overfitted (i.e. my model training can run for more epoch and get a slightly better results). You can also search <a href=\"https://www.kaggle.com/clavereu\" target=\"_blank\">@clavereu</a> solution on this forum, he also used <code>user_answer</code>.</p>",
      "rawMarkdown": "I did it, it helps my model training less overfitted (i.e. my model training can run for more epoch and get a slightly better results). You can also search @clavereu solution on this forum, he also used `user_answer`.",
      "votes": null
    },
    {
      "id": "1153413",
      "postDate": "01/14/2021 21:41:52",
      "content": "<p>We tried it too. Didn't help for us.</p>",
      "rawMarkdown": "We tried it too. Didn't help for us.",
      "votes": null
    },
    {
      "id": "1156603",
      "postDate": "01/17/2021 09:28:20",
      "content": "<p>Thanks for coming back! I'll check out <a href=\"https://www.kaggle.com/clavereu\" target=\"_blank\">@clavereu</a>'s solution. </p>",
      "rawMarkdown": "Thanks for coming back! I'll check out @clavereu's solution.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1153226,
      "author_name": "yihdarshieh",
      "author_url": "",
      "post_date": "01/14/2021 17:40:53",
      "content": "<p>I did it, it helps my model training less overfitted (i.e. my model training can run for more epoch and get a slightly better results). You can also search <a href=\"https://www.kaggle.com/clavereu\" target=\"_blank\">@clavereu</a> solution on this forum, he also used <code>user_answer</code>.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1153413,
      "author_name": "antorsae",
      "author_url": "",
      "post_date": "01/14/2021 21:41:52",
      "content": "<p>We tried it too. Didn't help for us.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1156603,
      "author_name": "nohitwonder",
      "author_url": "",
      "post_date": "01/17/2021 09:28:20",
      "content": "<p>Thanks for coming back! I'll check out <a href=\"https://www.kaggle.com/clavereu\" target=\"_blank\">@clavereu</a>'s solution. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1153123": "Hi, I am curious to hear about any attempts to set up and train auxillary models against actually given user answers as part of a greater solution?\n\nI did not find time to investigate this myself. It sounds 'less wrong' to opt for 'Stockholm' as the capital of Norway rather than 'Kuala Lumpur', and I am wondering if there might as such be signal in the actual value of 'user_answer'. Thanks!",
    "1153226": "I did it, it helps my model training less overfitted (i.e. my model training can run for more epoch and get a slightly better results). You can also search @clavereu solution on this forum, he also used `user_answer`.",
    "1153413": "We tried it too. Didn't help for us.",
    "1156603": "Thanks for coming back! I'll check out @clavereu's solution."
  },
  "source": "meta"
}