{
  "id": 95964,
  "title": "What do you think the final LB scores will get down to?",
  "url": "/competitions/champs-scalar-coupling/discussion/95964",
  "author_name": "",
  "post_date": "2019-06-16T21:17:47.502069800Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Being new to Kaggle I can't really gauge how these competitions progress... Have we seen the main approaches and now it's small incremental gains, maybe ending around -4?\nOr are there usually a few breakthroughs and we'll get down to -10? -15?\nIs it usually a steady flow of kernels, or a sudden flurry at the end?</p>\n\n<p>QM9 controversy aside, I'm really enjoying watching all the discussion and fascinating kernel ideas coming out. This \"open science\" approach where you publish as you go is definitely the future. Advances are made faster and the end result is of a higher quality.</p>",
  "messages": [
    {
      "id": "554041",
      "postDate": "06/16/2019 21:17:47",
      "content": "<p>Being new to Kaggle I can't really gauge how these competitions progress... Have we seen the main approaches and now it's small incremental gains, maybe ending around -4?\nOr are there usually a few breakthroughs and we'll get down to -10? -15?\nIs it usually a steady flow of kernels, or a sudden flurry at the end?</p>\n\n<p>QM9 controversy aside, I'm really enjoying watching all the discussion and fascinating kernel ideas coming out. This \"open science\" approach where you publish as you go is definitely the future. Advances are made faster and the end result is of a higher quality.</p>",
      "rawMarkdown": "Being new to Kaggle I can't really gauge how these competitions progress... Have we seen the main approaches and now it's small incremental gains, maybe ending around -4?\nOr are there usually a few breakthroughs and we'll get down to -10? -15?\nIs it usually a steady flow of kernels, or a sudden flurry at the end?\n\nQM9 controversy aside, I'm really enjoying watching all the discussion and fascinating kernel ideas coming out. This \"open science\" approach where you publish as you go is definitely the future. Advances are made faster and the end result is of a higher quality.",
      "votes": null
    },
    {
      "id": "554049",
      "postDate": "06/16/2019 22:13:14",
      "content": "<p>I've been wondering the same thing.  It's still early days.  I'm expecting that the leading scores at the end will be significantly lower.  There's lots and lots of room for feature engineering in this competition.</p>",
      "rawMarkdown": "I've been wondering the same thing.  It's still early days.  I'm expecting that the leading scores at the end will be significantly lower.  There's lots and lots of room for feature engineering in this competition.",
      "votes": null
    },
    {
      "id": "554454",
      "postDate": "06/17/2019 12:21:44",
      "content": "<p>Not necessarily. Remember Santander competition where it took 1 month to get 0.002 boost... And those who got higher after that (in the next 3 months) got medals.</p>",
      "rawMarkdown": "Not necessarily. Remember Santander competition where it took 1 month to get 0.002 boost... And those who got higher after that (in the next 3 months) got medals.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 554049,
      "author_name": "abendorf",
      "author_url": "",
      "post_date": "06/16/2019 22:13:14",
      "content": "<p>I've been wondering the same thing.  It's still early days.  I'm expecting that the leading scores at the end will be significantly lower.  There's lots and lots of room for feature engineering in this competition.</p>",
      "votes": null,
      "replies": [
        {
          "id": 554454,
          "author_name": "stanislavblinov",
          "author_url": "",
          "post_date": "06/17/2019 12:21:44",
          "content": "<p>Not necessarily. Remember Santander competition where it took 1 month to get 0.002 boost... And those who got higher after that (in the next 3 months) got medals.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "554041": "Being new to Kaggle I can't really gauge how these competitions progress... Have we seen the main approaches and now it's small incremental gains, maybe ending around -4?\nOr are there usually a few breakthroughs and we'll get down to -10? -15?\nIs it usually a steady flow of kernels, or a sudden flurry at the end?\n\nQM9 controversy aside, I'm really enjoying watching all the discussion and fascinating kernel ideas coming out. This \"open science\" approach where you publish as you go is definitely the future. Advances are made faster and the end result is of a higher quality.",
    "554049": "I've been wondering the same thing.  It's still early days.  I'm expecting that the leading scores at the end will be significantly lower.  There's lots and lots of room for feature engineering in this competition.",
    "554454": "Not necessarily. Remember Santander competition where it took 1 month to get 0.002 boost... And those who got higher after that (in the next 3 months) got medals."
  },
  "source": "meta"
}