{
  "id": 98722,
  "title": "How many features?",
  "url": "/competitions/champs-scalar-coupling/discussion/98722",
  "author_name": "",
  "post_date": "2019-07-05T19:42:09.829629500Z",
  "votes": null,
  "comment_count": 5,
  "views": 0,
  "content": "<p>So far, I have generated few hundreds of features and I am working on reducing this number while getting similar (or better) results. \nI am curious about the number of features you are using. :)</p>",
  "messages": [
    {
      "id": "568981",
      "postDate": "07/05/2019 19:42:09",
      "content": "<p>So far, I have generated few hundreds of features and I am working on reducing this number while getting similar (or better) results. \nI am curious about the number of features you are using. :)</p>",
      "rawMarkdown": "So far, I have generated few hundreds of features and I am working on reducing this number while getting similar (or better) results. \nI am curious about the number of features you are using. :)",
      "votes": null
    },
    {
      "id": "568986",
      "postDate": "07/05/2019 19:46:59",
      "content": "<p>All of them!\nTryed to bruteforce (800+ for small types) - that didnt end well..</p>",
      "rawMarkdown": "All of them!\nTryed to bruteforce (800+ for small types) - that didnt end well..",
      "votes": null
    },
    {
      "id": "569081",
      "postDate": "07/06/2019 02:54:59",
      "content": "<p>Hi, How have you generated the features ? I am wondering if there is any way to generate the additional information for the test data, that are available for training data.</p>",
      "rawMarkdown": "Hi, How have you generated the features ? I am wondering if there is any way to generate the additional information for the test data, that are available for training data.",
      "votes": null
    },
    {
      "id": "569132",
      "postDate": "07/06/2019 06:05:03",
      "content": "<p>Yes it is possible by training a model to predict the missing feature. Just make sure to predict on OOF to avoid overfitting. For the fc column, you can find few kernels. \nLet me know if you need more details. </p>",
      "rawMarkdown": "Yes it is possible by training a model to predict the missing feature. Just make sure to predict on OOF to avoid overfitting. For the fc column, you can find few kernels. \nLet me know if you need more details.",
      "votes": null
    },
    {
      "id": "569489",
      "postDate": "07/06/2019 18:48:38",
      "content": "<p>Less than 100, LB -1.02</p>",
      "rawMarkdown": "Less than 100, LB -1.02",
      "votes": null
    },
    {
      "id": "572148",
      "postDate": "07/10/2019 14:36:30",
      "content": "<p>~300 features: LB: ~-1.8. \n I think 300 for -1.8 is too many though</p>",
      "rawMarkdown": "~300 features: LB: ~-1.8. \n I think 300 for -1.8 is too many though",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 568986,
      "author_name": "stanislavblinov",
      "author_url": "",
      "post_date": "07/05/2019 19:46:59",
      "content": "<p>All of them!\nTryed to bruteforce (800+ for small types) - that didnt end well..</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 569081,
      "author_name": "manasghosal",
      "author_url": "",
      "post_date": "07/06/2019 02:54:59",
      "content": "<p>Hi, How have you generated the features ? I am wondering if there is any way to generate the additional information for the test data, that are available for training data.</p>",
      "votes": null,
      "replies": [
        {
          "id": 569132,
          "author_name": "yassinealouini",
          "author_url": "",
          "post_date": "07/06/2019 06:05:03",
          "content": "<p>Yes it is possible by training a model to predict the missing feature. Just make sure to predict on OOF to avoid overfitting. For the fc column, you can find few kernels. \nLet me know if you need more details. </p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 569489,
      "author_name": "saurabh7",
      "author_url": "",
      "post_date": "07/06/2019 18:48:38",
      "content": "<p>Less than 100, LB -1.02</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 572148,
      "author_name": "leptony",
      "author_url": "",
      "post_date": "07/10/2019 14:36:30",
      "content": "<p>~300 features: LB: ~-1.8. \n I think 300 for -1.8 is too many though</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "568981": "So far, I have generated few hundreds of features and I am working on reducing this number while getting similar (or better) results. \nI am curious about the number of features you are using. :)",
    "568986": "All of them!\nTryed to bruteforce (800+ for small types) - that didnt end well..",
    "569081": "Hi, How have you generated the features ? I am wondering if there is any way to generate the additional information for the test data, that are available for training data.",
    "569132": "Yes it is possible by training a model to predict the missing feature. Just make sure to predict on OOF to avoid overfitting. For the fc column, you can find few kernels. \nLet me know if you need more details.",
    "569489": "Less than 100, LB -1.02",
    "572148": "~300 features: LB: ~-1.8. \n I think 300 for -1.8 is too many though"
  },
  "source": "meta"
}