{
  "id": 292382,
  "title": "Num of Feature",
  "url": "/competitions/g-research-crypto-forecasting/discussion/292382",
  "author_name": "",
  "post_date": "2021-12-02T07:11:37.035739700Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I recently joined this competition.<br>\nNow, I am tackling feature engineering. So, I wondering how many feature everyone use? <br>\nCould I get your opinion?</p>\n<p>My current feature num is about 40. I will make more feature.<br>\n(At the same time, though, we have to tackle speed and memory issues.)</p>",
  "messages": [
    {
      "id": "1602863",
      "postDate": "12/02/2021 07:11:37",
      "content": "<p>I recently joined this competition.<br>\nNow, I am tackling feature engineering. So, I wondering how many feature everyone use? <br>\nCould I get your opinion?</p>\n<p>My current feature num is about 40. I will make more feature.<br>\n(At the same time, though, we have to tackle speed and memory issues.)</p>",
      "rawMarkdown": "I recently joined this competition.\nNow, I am tackling feature engineering. So, I wondering how many feature everyone use? \nCould I get your opinion?\n\nMy current feature num is about 40. I will make more feature.\n(At the same time, though, we have to tackle speed and memory issues.)",
      "votes": null
    },
    {
      "id": "1603451",
      "postDate": "12/02/2021 14:37:14",
      "content": "<p>The less, the better: It is easy to overfit this data</p>",
      "rawMarkdown": "The less, the better: It is easy to overfit this data",
      "votes": null
    },
    {
      "id": "1616333",
      "postDate": "12/13/2021 10:54:51",
      "content": "<p>It's very easy to get scoring error or timeout in this competition, prob bc of the API. :/</p>",
      "rawMarkdown": "It's very easy to get scoring error or timeout in this competition, prob bc of the API. :/",
      "votes": null
    },
    {
      "id": "1618549",
      "postDate": "12/15/2021 04:59:12",
      "content": "<p>It's very easy to overfit in this competition, especially if you use non-linear models. But you won't get the most out of linear models.</p>\n<p>You want to consider as many candidate features as crosses your mind, but wrap your non-linear regressor around a principled feature selection approach. </p>\n<p>I wrote a series of notebooks on this: <a href=\"https://www.kaggle.com/leanboosting/automating-feature-engineering-part-i\" target=\"_blank\">Part I</a>, <a href=\"https://www.kaggle.com/leanboosting/automating-feature-engineering-part-ii\" target=\"_blank\">Part II</a>, <a href=\"https://www.kaggle.com/leanboosting/automating-feature-engineering-part-iii\" target=\"_blank\">Part III</a>.</p>",
      "rawMarkdown": "It's very easy to overfit in this competition, especially if you use non-linear models. But you won't get the most out of linear models.\n\nYou want to consider as many candidate features as crosses your mind, but wrap your non-linear regressor around a principled feature selection approach. \n\nI wrote a series of notebooks on this: [Part I](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-i), [Part II](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-ii), [Part III](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-iii).",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1603451,
      "author_name": "",
      "author_url": "",
      "post_date": "12/02/2021 14:37:14",
      "content": "<p>The less, the better: It is easy to overfit this data</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1616333,
      "author_name": "carol916",
      "author_url": "",
      "post_date": "12/13/2021 10:54:51",
      "content": "<p>It's very easy to get scoring error or timeout in this competition, prob bc of the API. :/</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1618549,
      "author_name": "leanboosting",
      "author_url": "",
      "post_date": "12/15/2021 04:59:12",
      "content": "<p>It's very easy to overfit in this competition, especially if you use non-linear models. But you won't get the most out of linear models.</p>\n<p>You want to consider as many candidate features as crosses your mind, but wrap your non-linear regressor around a principled feature selection approach. </p>\n<p>I wrote a series of notebooks on this: <a href=\"https://www.kaggle.com/leanboosting/automating-feature-engineering-part-i\" target=\"_blank\">Part I</a>, <a href=\"https://www.kaggle.com/leanboosting/automating-feature-engineering-part-ii\" target=\"_blank\">Part II</a>, <a href=\"https://www.kaggle.com/leanboosting/automating-feature-engineering-part-iii\" target=\"_blank\">Part III</a>.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1602863": "I recently joined this competition.\nNow, I am tackling feature engineering. So, I wondering how many feature everyone use? \nCould I get your opinion?\n\nMy current feature num is about 40. I will make more feature.\n(At the same time, though, we have to tackle speed and memory issues.)",
    "1603451": "The less, the better: It is easy to overfit this data",
    "1616333": "It's very easy to get scoring error or timeout in this competition, prob bc of the API. :/",
    "1618549": "It's very easy to overfit in this competition, especially if you use non-linear models. But you won't get the most out of linear models.\n\nYou want to consider as many candidate features as crosses your mind, but wrap your non-linear regressor around a principled feature selection approach. \n\nI wrote a series of notebooks on this: [Part I](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-i), [Part II](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-ii), [Part III](https://www.kaggle.com/leanboosting/automating-feature-engineering-part-iii)."
  },
  "source": "meta"
}