{
  "id": 55587,
  "title": "Validation Approaches",
  "url": "/competitions/avito-demand-prediction/discussion/55587",
  "author_name": "",
  "post_date": "2018-04-29T05:25:49.428723600Z",
  "votes": 4,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello, \nI have been  messing with  cross-validation aproaches. I am currently using a simple 5-fold, but my validation RMSE does not seem correlated with  leaderboard at all. What are you guys using? In SRK kernel he used the last rows of the train set as validation, but I don't see why that would make more sense than  fold validation.</p>",
  "messages": [
    {
      "id": "320573",
      "postDate": "04/29/2018 05:25:49",
      "content": "<p>Hello, \nI have been  messing with  cross-validation aproaches. I am currently using a simple 5-fold, but my validation RMSE does not seem correlated with  leaderboard at all. What are you guys using? In SRK kernel he used the last rows of the train set as validation, but I don't see why that would make more sense than  fold validation.</p>",
      "rawMarkdown": "Hello, \nI have been  messing with  cross-validation aproaches. I am currently using a simple 5-fold, but my validation RMSE does not seem correlated with  leaderboard at all. What are you guys using? In SRK kernel he used the last rows of the train set as validation, but I don't see why that would make more sense than  fold validation.",
      "votes": null
    },
    {
      "id": "320605",
      "postDate": "04/29/2018 08:43:58",
      "content": "<p>For me 5-fold validation with RMSE follows my score on leaderboard.</p>",
      "rawMarkdown": "For me 5-fold validation with RMSE follows my score on leaderboard.",
      "votes": null
    },
    {
      "id": "320780",
      "postDate": "04/29/2018 20:15:56",
      "content": "<p>I've been using activation dates &gt;= 03-26 and my validation improvement seems to move with LB.</p>",
      "rawMarkdown": "I've been using activation dates &gt;= 03-26 and my validation improvement seems to move with LB.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 320605,
      "author_name": "artgor",
      "author_url": "",
      "post_date": "04/29/2018 08:43:58",
      "content": "<p>For me 5-fold validation with RMSE follows my score on leaderboard.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 320780,
      "author_name": "brandenkmurray",
      "author_url": "",
      "post_date": "04/29/2018 20:15:56",
      "content": "<p>I've been using activation dates &gt;= 03-26 and my validation improvement seems to move with LB.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "320573": "Hello, \nI have been  messing with  cross-validation aproaches. I am currently using a simple 5-fold, but my validation RMSE does not seem correlated with  leaderboard at all. What are you guys using? In SRK kernel he used the last rows of the train set as validation, but I don't see why that would make more sense than  fold validation.",
    "320605": "For me 5-fold validation with RMSE follows my score on leaderboard.",
    "320780": "I've been using activation dates &gt;= 03-26 and my validation improvement seems to move with LB."
  },
  "source": "meta"
}