{
  "id": 300127,
  "title": "Should LSTM be Used? If Yes, How?",
  "url": "/competitions/g-research-crypto-forecasting/discussion/300127",
  "author_name": "samuel adeshina",
  "post_date": "2022-01-11T08:46:53.209000",
  "votes": 1,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi Kagglers,</p>\n<p>I have been thinking of using LSTM for this problem. However, I have some concerns: we have data on 14 assets. If we choose to predict what we have on the 31st minute, say, then we will need the previous 30-minute data. These 30-minute data are data on 14 assets but the 31st minute data are data on just one asset. This means we will be predicting data on one asset using data on 14 assets. This does not make sense to me.</p>\n<p>Also, if we train the first 30 rows and try to use it to predict the data on the 31st row, the 31st row may share the same time with the 30th row. I think this is also a problem.</p>\n<p>How should we solve these problems?<br>\nThanks.</p>",
  "messages": [
    {
      "id": 1645811,
      "postDate": "2022-01-11T08:46:53.210Z",
      "content": "<p>Hi Kagglers,</p>\n<p>I have been thinking of using LSTM for this problem. However, I have some concerns: we have data on 14 assets. If we choose to predict what we have on the 31st minute, say, then we will need the previous 30-minute data. These 30-minute data are data on 14 assets but the 31st minute data are data on just one asset. This means we will be predicting data on one asset using data on 14 assets. This does not make sense to me.</p>\n<p>Also, if we train the first 30 rows and try to use it to predict the data on the 31st row, the 31st row may share the same time with the 30th row. I think this is also a problem.</p>\n<p>How should we solve these problems?<br>\nThanks.</p>",
      "rawMarkdown": "Hi Kagglers,\n\nI have been thinking of using LSTM for this problem. However, I have some concerns: we have data on 14 assets. If we choose to predict what we have on the 31st minute, say, then we will need the previous 30-minute data. These 30-minute data are data on 14 assets but the 31st minute data are data on just one asset. This means we will be predicting data on one asset using data on 14 assets. This does not make sense to me.\n\nAlso, if we train the first 30 rows and try to use it to predict the data on the 31st row, the 31st row may share the same time with the 30th row. I think this is also a problem.\n\nHow should we solve these problems?\nThanks.",
      "votes": 1
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1645811": "Hi Kagglers,\n\nI have been thinking of using LSTM for this problem. However, I have some concerns: we have data on 14 assets. If we choose to predict what we have on the 31st minute, say, then we will need the previous 30-minute data. These 30-minute data are data on 14 assets but the 31st minute data are data on just one asset. This means we will be predicting data on one asset using data on 14 assets. This does not make sense to me.\n\nAlso, if we train the first 30 rows and try to use it to predict the data on the 31st row, the 31st row may share the same time with the 30th row. I think this is also a problem.\n\nHow should we solve these problems?\nThanks."
  }
}