{
  "id": 129551,
  "title": "AS for adversarial validation, this is serious problem",
  "url": "/competitions/cat-in-the-dat-ii/discussion/129551",
  "author_name": "",
  "post_date": "2020-02-08T21:30:52.134942400Z",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>AS for adversarial validation, it was shown that auc is about 0.65 by some kaggler.\n(<a href=\"https://www.kaggle.com/tunguz/adversicat-ii\">https://www.kaggle.com/tunguz/adversicat-ii</a>).</p>\n\n<p>But,    I tried one_hot noms, together with other features, auc is around 0.5.</p>\n\n<p>And, target_encoded noms, together with other features, auc is around 0.999.</p>\n\n<p>May it suggest that target_encoding ruin homogeneity of train set and test set, hence, cv score is different from lb score?</p>\n\n<p>Anyone knows the reason?</p>\n\n<p>THANKS!</p>\n\n<p>PS:</p>\n\n<p>I thought over and over about this question, and figured a trick out ,thus ,pb score increased by 0.00011, from 0.78600 to 0.78611.</p>",
  "messages": [
    {
      "id": "740101",
      "postDate": "02/08/2020 21:30:52",
      "content": "<p>AS for adversarial validation, it was shown that auc is about 0.65 by some kaggler.\n(<a href=\"https://www.kaggle.com/tunguz/adversicat-ii\">https://www.kaggle.com/tunguz/adversicat-ii</a>).</p>\n\n<p>But,    I tried one_hot noms, together with other features, auc is around 0.5.</p>\n\n<p>And, target_encoded noms, together with other features, auc is around 0.999.</p>\n\n<p>May it suggest that target_encoding ruin homogeneity of train set and test set, hence, cv score is different from lb score?</p>\n\n<p>Anyone knows the reason?</p>\n\n<p>THANKS!</p>\n\n<p>PS:</p>\n\n<p>I thought over and over about this question, and figured a trick out ,thus ,pb score increased by 0.00011, from 0.78600 to 0.78611.</p>",
      "rawMarkdown": "AS for adversarial validation, it was shown that auc is about 0.65 by some kaggler.\n(https://www.kaggle.com/tunguz/adversicat-ii).\n\nBut,    I tried one_hot noms, together with other features, auc is around 0.5.\n\nAnd, target_encoded noms, together with other features, auc is around 0.999.\n\nMay it suggest that target_encoding ruin homogeneity of train set and test set, hence, cv score is different from lb score?\n\nAnyone knows the reason?\n\nTHANKS!\n\nPS:\n\nI thought over and over about this question, and figured a trick out ,thus ,pb score increased by 0.00011, from 0.78600 to 0.78611.",
      "votes": null
    },
    {
      "id": "740130",
      "postDate": "02/08/2020 23:39:33",
      "content": "<p>Target encoding is a prediction. Therefore, it is common to use mean smoothing or noise to decrease prediction variance (overfitting).</p>",
      "rawMarkdown": "Target encoding is a prediction. Therefore, it is common to use mean smoothing or noise to decrease prediction variance (overfitting).",
      "votes": null
    },
    {
      "id": "760504",
      "postDate": "03/01/2020 11:34:15",
      "content": "<p><a href=\"/kailai\">@kailai</a>  </p>\n\n<blockquote>\n  <p>I thought over and over about this question, and figured a trick out </p>\n</blockquote>\n\n<p>I am faced same problem. So could you tell me this problem, please.</p>",
      "rawMarkdown": "kailai  \n&gt;I thought over and over about this question, and figured a trick out \n\n I am faced same problem. So could you tell me this problem, please.",
      "votes": null
    },
    {
      "id": "786938",
      "postDate": "03/26/2020 11:16:17",
      "content": "<p>Does anyone use the adversarial validation in this challenge?</p>",
      "rawMarkdown": "Does anyone use the adversarial validation in this challenge?",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 740130,
      "author_name": "frankmollard",
      "author_url": "",
      "post_date": "02/08/2020 23:39:33",
      "content": "<p>Target encoding is a prediction. Therefore, it is common to use mean smoothing or noise to decrease prediction variance (overfitting).</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 760504,
      "author_name": "tenajima",
      "author_url": "",
      "post_date": "03/01/2020 11:34:15",
      "content": "<p><a href=\"/kailai\">@kailai</a>  </p>\n\n<blockquote>\n  <p>I thought over and over about this question, and figured a trick out </p>\n</blockquote>\n\n<p>I am faced same problem. So could you tell me this problem, please.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 786938,
      "author_name": "vcerchez",
      "author_url": "",
      "post_date": "03/26/2020 11:16:17",
      "content": "<p>Does anyone use the adversarial validation in this challenge?</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "740101": "AS for adversarial validation, it was shown that auc is about 0.65 by some kaggler.\n(https://www.kaggle.com/tunguz/adversicat-ii).\n\nBut,    I tried one_hot noms, together with other features, auc is around 0.5.\n\nAnd, target_encoded noms, together with other features, auc is around 0.999.\n\nMay it suggest that target_encoding ruin homogeneity of train set and test set, hence, cv score is different from lb score?\n\nAnyone knows the reason?\n\nTHANKS!\n\nPS:\n\nI thought over and over about this question, and figured a trick out ,thus ,pb score increased by 0.00011, from 0.78600 to 0.78611.",
    "740130": "Target encoding is a prediction. Therefore, it is common to use mean smoothing or noise to decrease prediction variance (overfitting).",
    "760504": "kailai  \n&gt;I thought over and over about this question, and figured a trick out \n\n I am faced same problem. So could you tell me this problem, please.",
    "786938": "Does anyone use the adversarial validation in this challenge?"
  },
  "source": "meta"
}