{
  "id": 46305,
  "title": "Surprised to see only few Time Series kernels",
  "url": "/competitions/favorita-grocery-sales-forecasting/discussion/46305",
  "author_name": "",
  "post_date": "2017-12-24T11:40:11.839460200Z",
  "votes": 4,
  "comment_count": 3,
  "views": 0,
  "content": "<p>I was wondering why very less number of Time Series kernels are there in the kernels section. I am just starting the competition. Is there anything i am missing. Is time series analysis a wrong way to approach the problem?</p>",
  "messages": [
    {
      "id": "261885",
      "postDate": "12/24/2017 11:40:11",
      "content": "<p>I was wondering why very less number of Time Series kernels are there in the kernels section. I am just starting the competition. Is there anything i am missing. Is time series analysis a wrong way to approach the problem?</p>",
      "rawMarkdown": "I was wondering why very less number of Time Series kernels are there in the kernels section. I am just starting the competition. Is there anything i am missing. Is time series analysis a wrong way to approach the problem?",
      "votes": null
    },
    {
      "id": "262010",
      "postDate": "12/24/2017 22:29:04",
      "content": "<p>The MLs used here are structured in the way Time Series are, so maybe someone is using that kind of approach but just not publishing it.</p>",
      "rawMarkdown": "The MLs used here are structured in the way Time Series are, so maybe someone is using that kind of approach but just not publishing it.",
      "votes": null
    },
    {
      "id": "263354",
      "postDate": "12/29/2017 20:48:35",
      "content": "<p>If you're using lagged features, moving averages, etc., etc. with a gradient boosting regression or LSTM, then you're basically doing something in the same spirit of sARIMA or GARCH modeling. But why use a \"data first\" approach when you can use a \"model first\" approach?</p>",
      "rawMarkdown": "If you're using lagged features, moving averages, etc., etc. with a gradient boosting regression or LSTM, then you're basically doing something in the same spirit of sARIMA or GARCH modeling. But why use a \"data first\" approach when you can use a \"model first\" approach?",
      "votes": null
    },
    {
      "id": "264053",
      "postDate": "01/02/2018 04:49:05",
      "content": "<p>I only ues LGB</p>",
      "rawMarkdown": "I only ues LGB",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 262010,
      "author_name": "showletter",
      "author_url": "",
      "post_date": "12/24/2017 22:29:04",
      "content": "<p>The MLs used here are structured in the way Time Series are, so maybe someone is using that kind of approach but just not publishing it.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 263354,
      "author_name": "puremath86",
      "author_url": "",
      "post_date": "12/29/2017 20:48:35",
      "content": "<p>If you're using lagged features, moving averages, etc., etc. with a gradient boosting regression or LSTM, then you're basically doing something in the same spirit of sARIMA or GARCH modeling. But why use a \"data first\" approach when you can use a \"model first\" approach?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 264053,
      "author_name": "xuanhahaha",
      "author_url": "",
      "post_date": "01/02/2018 04:49:05",
      "content": "<p>I only ues LGB</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "261885": "I was wondering why very less number of Time Series kernels are there in the kernels section. I am just starting the competition. Is there anything i am missing. Is time series analysis a wrong way to approach the problem?",
    "262010": "The MLs used here are structured in the way Time Series are, so maybe someone is using that kind of approach but just not publishing it.",
    "263354": "If you're using lagged features, moving averages, etc., etc. with a gradient boosting regression or LSTM, then you're basically doing something in the same spirit of sARIMA or GARCH modeling. But why use a \"data first\" approach when you can use a \"model first\" approach?",
    "264053": "I only ues LGB"
  },
  "source": "meta"
}