{
  "id": 53257,
  "title": "Adversarial Validation",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/53257",
  "author_name": "",
  "post_date": "2018-03-28T16:19:35.380143400Z",
  "votes": 6,
  "comment_count": 2,
  "views": 0,
  "content": "<p>A lot of effort(kernel and discussion) was being put into finding if train and test set(except time based features) are similar or not. That's why I decided to find what Adversarial Validation approach has to say about it. Below are the results of two runs(with and without \"IP\" feature).</p>\n\n<pre><code>Fold 0 Val Auc:0.706425\nFold 1 Val Auc:0.706085\nFold 2 Val Auc:0.706530\nFold 3 Val Auc:0.706761\nFold 4 Val Auc:0.706397\n</code></pre>\n\n<p>Features used in above run = ['ip','app', 'os', 'channel', 'device']</p>\n\n<pre><code>Fold 0 Val Auc:0.636470\nFold 1 Val Auc:0.636752\nFold 2 Val Auc:0.636799\nFold 3 Val Auc:0.636959\nFold 4 Val Auc:0.636957\n</code></pre>\n\n<p>Features used in above run =  [''app', 'os', 'channel', 'device']</p>\n\n<p><a href=\"https://www.kaggle.com/sohaibomar/adversarial-validation\">Here's</a> the link to my kernel. Please share your views about the results.\nI have recently learned about Advarsarial Validation from @olivier and @konard in Toxic competition if I am missing something then feel free to correct me.</p>\n\n<p>Thanks</p>",
  "messages": [
    {
      "id": "305236",
      "postDate": "03/28/2018 16:19:35",
      "content": "<p>A lot of effort(kernel and discussion) was being put into finding if train and test set(except time based features) are similar or not. That's why I decided to find what Adversarial Validation approach has to say about it. Below are the results of two runs(with and without \"IP\" feature).</p>\n\n<pre><code>Fold 0 Val Auc:0.706425\nFold 1 Val Auc:0.706085\nFold 2 Val Auc:0.706530\nFold 3 Val Auc:0.706761\nFold 4 Val Auc:0.706397\n</code></pre>\n\n<p>Features used in above run = ['ip','app', 'os', 'channel', 'device']</p>\n\n<pre><code>Fold 0 Val Auc:0.636470\nFold 1 Val Auc:0.636752\nFold 2 Val Auc:0.636799\nFold 3 Val Auc:0.636959\nFold 4 Val Auc:0.636957\n</code></pre>\n\n<p>Features used in above run =  [''app', 'os', 'channel', 'device']</p>\n\n<p><a href=\"https://www.kaggle.com/sohaibomar/adversarial-validation\">Here's</a> the link to my kernel. Please share your views about the results.\nI have recently learned about Advarsarial Validation from @olivier and @konard in Toxic competition if I am missing something then feel free to correct me.</p>\n\n<p>Thanks</p>",
      "rawMarkdown": "A lot of effort(kernel and discussion) was being put into finding if train and test set(except time based features) are similar or not. That's why I decided to find what Adversarial Validation approach has to say about it. Below are the results of two runs(with and without \"IP\" feature).\n\n    Fold 0 Val Auc:0.706425\n    Fold 1 Val Auc:0.706085\n    Fold 2 Val Auc:0.706530\n    Fold 3 Val Auc:0.706761\n    Fold 4 Val Auc:0.706397\n\nFeatures used in above run = ['ip','app', 'os', 'channel', 'device']\n\n    Fold 0 Val Auc:0.636470\n    Fold 1 Val Auc:0.636752\n    Fold 2 Val Auc:0.636799\n    Fold 3 Val Auc:0.636959\n    Fold 4 Val Auc:0.636957\n\nFeatures used in above run =  [''app', 'os', 'channel', 'device']\n\n[Here's][1] the link to my kernel. Please share your views about the results.\nI have recently learned about Advarsarial Validation from @olivier and @konard in Toxic competition if I am missing something then feel free to correct me.\n\nThanks\n\n\n  [1]: https://www.kaggle.com/sohaibomar/adversarial-validation",
      "votes": null
    },
    {
      "id": "307250",
      "postDate": "04/01/2018 07:21:53",
      "content": "<p>Thank you a lot for sharing this @Sohaib ! Something I can learn from!</p>",
      "rawMarkdown": "Thank you a lot for sharing this @Sohaib ! Something I can learn from!",
      "votes": null
    },
    {
      "id": "307492",
      "postDate": "04/01/2018 19:52:46",
      "content": "<p>Thanks, I am glad it helped you learn. </p>",
      "rawMarkdown": "Thanks, I am glad it helped you learn.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 307250,
      "author_name": "asparuhhristov",
      "author_url": "",
      "post_date": "04/01/2018 07:21:53",
      "content": "<p>Thank you a lot for sharing this @Sohaib ! Something I can learn from!</p>",
      "votes": null,
      "replies": [
        {
          "id": 307492,
          "author_name": "sohaibomar",
          "author_url": "",
          "post_date": "04/01/2018 19:52:46",
          "content": "<p>Thanks, I am glad it helped you learn. </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "305236": "A lot of effort(kernel and discussion) was being put into finding if train and test set(except time based features) are similar or not. That's why I decided to find what Adversarial Validation approach has to say about it. Below are the results of two runs(with and without \"IP\" feature).\n\n    Fold 0 Val Auc:0.706425\n    Fold 1 Val Auc:0.706085\n    Fold 2 Val Auc:0.706530\n    Fold 3 Val Auc:0.706761\n    Fold 4 Val Auc:0.706397\n\nFeatures used in above run = ['ip','app', 'os', 'channel', 'device']\n\n    Fold 0 Val Auc:0.636470\n    Fold 1 Val Auc:0.636752\n    Fold 2 Val Auc:0.636799\n    Fold 3 Val Auc:0.636959\n    Fold 4 Val Auc:0.636957\n\nFeatures used in above run =  [''app', 'os', 'channel', 'device']\n\n[Here's][1] the link to my kernel. Please share your views about the results.\nI have recently learned about Advarsarial Validation from @olivier and @konard in Toxic competition if I am missing something then feel free to correct me.\n\nThanks\n\n\n  [1]: https://www.kaggle.com/sohaibomar/adversarial-validation",
    "307250": "Thank you a lot for sharing this @Sohaib ! Something I can learn from!",
    "307492": "Thanks, I am glad it helped you learn."
  },
  "source": "meta"
}