{
  "id": 53421,
  "title": "attributed_time?",
  "url": "/competitions/talkingdata-adtracking-fraud-detection/discussion/53421",
  "author_name": "",
  "post_date": "2018-03-30T11:30:25.082658900Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Does anyone have any idea how to use attributed time or should it be thrown away like most of the kernals?</p>",
  "messages": [
    {
      "id": "306400",
      "postDate": "03/30/2018 11:30:25",
      "content": "<p>Does anyone have any idea how to use attributed time or should it be thrown away like most of the kernals?</p>",
      "rawMarkdown": "Does anyone have any idea how to use attributed time or should it be thrown away like most of the kernals?",
      "votes": null
    },
    {
      "id": "306449",
      "postDate": "03/30/2018 12:48:04",
      "content": "<p>Pradeep G <a href=\"https://www.kaggle.com/gpradk/keras-starter-nn-with-embeddings\">uses it</a> to help extract embeddings. That's a special case of the general idea that it can be used to identify latent variables that might be useful for prediction. There are several techniques for doing this. Typically some variation on: fit a model to predict attributed time and use the predictions of that model as features to predict is_attributed. I haven't tried this, but it seems like a plausible approach.</p>",
      "rawMarkdown": "Pradeep G [uses it][1] to help extract embeddings. That's a special case of the general idea that it can be used to identify latent variables that might be useful for prediction. There are several techniques for doing this. Typically some variation on: fit a model to predict attributed time and use the predictions of that model as features to predict is_attributed. I haven't tried this, but it seems like a plausible approach.\n\n [1]: https://www.kaggle.com/gpradk/keras-starter-nn-with-embeddings",
      "votes": null
    },
    {
      "id": "307248",
      "postDate": "04/01/2018 07:14:00",
      "content": "<p>I haven't tried to use it and it is at the very bottom part of my to-do list for this competition. I really don't see a logical explanation why it would work out, but for anybody that does see - go ahead and give it a try!</p>",
      "rawMarkdown": "I haven't tried to use it and it is at the very bottom part of my to-do list for this competition. I really don't see a logical explanation why it would work out, but for anybody that does see - go ahead and give it a try!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 306449,
      "author_name": "aharless",
      "author_url": "",
      "post_date": "03/30/2018 12:48:04",
      "content": "<p>Pradeep G <a href=\"https://www.kaggle.com/gpradk/keras-starter-nn-with-embeddings\">uses it</a> to help extract embeddings. That's a special case of the general idea that it can be used to identify latent variables that might be useful for prediction. There are several techniques for doing this. Typically some variation on: fit a model to predict attributed time and use the predictions of that model as features to predict is_attributed. I haven't tried this, but it seems like a plausible approach.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 307248,
      "author_name": "asparuhhristov",
      "author_url": "",
      "post_date": "04/01/2018 07:14:00",
      "content": "<p>I haven't tried to use it and it is at the very bottom part of my to-do list for this competition. I really don't see a logical explanation why it would work out, but for anybody that does see - go ahead and give it a try!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "306400": "Does anyone have any idea how to use attributed time or should it be thrown away like most of the kernals?",
    "306449": "Pradeep G [uses it][1] to help extract embeddings. That's a special case of the general idea that it can be used to identify latent variables that might be useful for prediction. There are several techniques for doing this. Typically some variation on: fit a model to predict attributed time and use the predictions of that model as features to predict is_attributed. I haven't tried this, but it seems like a plausible approach.\n\n [1]: https://www.kaggle.com/gpradk/keras-starter-nn-with-embeddings",
    "307248": "I haven't tried to use it and it is at the very bottom part of my to-do list for this competition. I really don't see a logical explanation why it would work out, but for anybody that does see - go ahead and give it a try!"
  },
  "source": "meta"
}