{
  "id": 90799,
  "title": "Are we exploring temporal bins deep enough?",
  "url": "/competitions/LANL-Earthquake-Prediction/discussion/90799",
  "author_name": "",
  "post_date": "2019-04-27T11:43:40.403501700Z",
  "votes": 1,
  "comment_count": 2,
  "views": 0,
  "content": "<p>One thing I've been noticing in some public kernels is that we are using key segments to calculate specific features (for example, using the first 5k/10k points to calculate mean, std, ...). Is this enough? \nIt seems that we can benefit the most if we create a <strong>systematic approach</strong>, dividing the segments in subsegments. Any thoughts on this matter? Sugestions are welcome! 👍 \nNice work everyone.</p>",
  "messages": [
    {
      "id": "523918",
      "postDate": "04/27/2019 11:43:40",
      "content": "<p>One thing I've been noticing in some public kernels is that we are using key segments to calculate specific features (for example, using the first 5k/10k points to calculate mean, std, ...). Is this enough? \nIt seems that we can benefit the most if we create a <strong>systematic approach</strong>, dividing the segments in subsegments. Any thoughts on this matter? Sugestions are welcome! 👍 \nNice work everyone.</p>",
      "rawMarkdown": "One thing I've been noticing in some public kernels is that we are using key segments to calculate specific features (for example, using the first 5k/10k points to calculate mean, std, ...). Is this enough? \nIt seems that we can benefit the most if we create a **systematic approach**, dividing the segments in subsegments. Any thoughts on this matter? Sugestions are welcome! 👍 \nNice work everyone.",
      "votes": null
    },
    {
      "id": "524003",
      "postDate": "04/27/2019 16:34:05",
      "content": "<p>I had a similar idea early in the competition to build a hierarchical model (i.e. a model built on predictions of daughter models trained on sub-segments of the original segment). I gave up on the idea pretty fast after disappointing results. Maybe I did it wrong tho..</p>",
      "rawMarkdown": "I had a similar idea early in the competition to build a hierarchical model (i.e. a model built on predictions of daughter models trained on sub-segments of the original segment). I gave up on the idea pretty fast after disappointing results. Maybe I did it wrong tho..",
      "votes": null
    },
    {
      "id": "524645",
      "postDate": "04/29/2019 09:05:31",
      "content": "<p>CNN can do this for u</p>",
      "rawMarkdown": "CNN can do this for u",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 524003,
      "author_name": "amjad85",
      "author_url": "",
      "post_date": "04/27/2019 16:34:05",
      "content": "<p>I had a similar idea early in the competition to build a hierarchical model (i.e. a model built on predictions of daughter models trained on sub-segments of the original segment). I gave up on the idea pretty fast after disappointing results. Maybe I did it wrong tho..</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 524645,
      "author_name": "wlmike",
      "author_url": "",
      "post_date": "04/29/2019 09:05:31",
      "content": "<p>CNN can do this for u</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "523918": "One thing I've been noticing in some public kernels is that we are using key segments to calculate specific features (for example, using the first 5k/10k points to calculate mean, std, ...). Is this enough? \nIt seems that we can benefit the most if we create a **systematic approach**, dividing the segments in subsegments. Any thoughts on this matter? Sugestions are welcome! 👍 \nNice work everyone.",
    "524003": "I had a similar idea early in the competition to build a hierarchical model (i.e. a model built on predictions of daughter models trained on sub-segments of the original segment). I gave up on the idea pretty fast after disappointing results. Maybe I did it wrong tho..",
    "524645": "CNN can do this for u"
  },
  "source": "meta"
}