{
  "id": 46709,
  "title": "moving average method",
  "url": "/competitions/favorita-grocery-sales-forecasting/discussion/46709",
  "author_name": "Zhongheng Yang",
  "post_date": "2018-01-02T04:25:18.183000",
  "votes": 0,
  "comment_count": 0,
  "views": 0,
  "content": "<p>hi \n  When the kernels say about using \"moving average\",  I feel like those kernels only calculate the average, while I think moving average should be the \"nearby dates' average sale\" for each date of each store's every item, in order to avoid the noise. And then we may create the \"moving average\" feature for each date of each store's every item. \n  But that  \"nearby dates' average sale\"  feature, which I am coding on my PC, is too slow to generate due to too much data.\n  I feel like to deal with this time series problem, more time-series methods should be discussed. Anyone tried some  model that is specific for time-series problem?</p>",
  "messages": [
    {
      "id": 264045,
      "postDate": "2018-01-02T04:25:18.183Z",
      "content": "<p>hi \n  When the kernels say about using \"moving average\",  I feel like those kernels only calculate the average, while I think moving average should be the \"nearby dates' average sale\" for each date of each store's every item, in order to avoid the noise. And then we may create the \"moving average\" feature for each date of each store's every item. \n  But that  \"nearby dates' average sale\"  feature, which I am coding on my PC, is too slow to generate due to too much data.\n  I feel like to deal with this time series problem, more time-series methods should be discussed. Anyone tried some  model that is specific for time-series problem?</p>",
      "rawMarkdown": "hi \n  When the kernels say about using \"moving average\",  I feel like those kernels only calculate the average, while I think moving average should be the \"nearby dates' average sale\" for each date of each store's every item, in order to avoid the noise. And then we may create the \"moving average\" feature for each date of each store's every item. \n  But that  \"nearby dates' average sale\"  feature, which I am coding on my PC, is too slow to generate due to too much data.\n  I feel like to deal with this time series problem, more time-series methods should be discussed. Anyone tried some  model that is specific for time-series problem?"
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "264045": "hi \n  When the kernels say about using \"moving average\",  I feel like those kernels only calculate the average, while I think moving average should be the \"nearby dates' average sale\" for each date of each store's every item, in order to avoid the noise. And then we may create the \"moving average\" feature for each date of each store's every item. \n  But that  \"nearby dates' average sale\"  feature, which I am coding on my PC, is too slow to generate due to too much data.\n  I feel like to deal with this time series problem, more time-series methods should be discussed. Anyone tried some  model that is specific for time-series problem?"
  }
}