{
  "id": 117304,
  "title": "Time Series prediction (ARIMA, LSTM)",
  "url": "/competitions/ashrae-energy-prediction/discussion/117304",
  "author_name": "",
  "post_date": "2019-11-14T14:45:38.152334900Z",
  "votes": 3,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi, has anyone tried out predicting using ARIMA or LSTM? It seems that people are just using lightgbm with little consideration that it is a continuous time series problem. \nI'm quite new to machine learning and especially time series prediction, so i have a couple (maybe silly) questions regarding this:\n1. If using ARIMA, are you supposed to do rolling forecasting? if so, it seems the amount of training needed is crazy given how many data points there are. Also how are you supposed to do rolling forecasting with test set? based on lightgbm's predictions?\n2. How does ARIMA with external regressors work? and do these regressors have to be time series as well (temperature etc)? \n3. Is LSTM a good choice for this problem? has anyone tried that?\n4. It was mentioned in another discussion that you can do fourier and ARIMA, how does that work? \nsorry for the amateur questions, any advice is greatly appreciated! </p>",
  "messages": [
    {
      "id": "673122",
      "postDate": "11/14/2019 14:45:38",
      "content": "<p>Hi, has anyone tried out predicting using ARIMA or LSTM? It seems that people are just using lightgbm with little consideration that it is a continuous time series problem. \nI'm quite new to machine learning and especially time series prediction, so i have a couple (maybe silly) questions regarding this:\n1. If using ARIMA, are you supposed to do rolling forecasting? if so, it seems the amount of training needed is crazy given how many data points there are. Also how are you supposed to do rolling forecasting with test set? based on lightgbm's predictions?\n2. How does ARIMA with external regressors work? and do these regressors have to be time series as well (temperature etc)? \n3. Is LSTM a good choice for this problem? has anyone tried that?\n4. It was mentioned in another discussion that you can do fourier and ARIMA, how does that work? \nsorry for the amateur questions, any advice is greatly appreciated! </p>",
      "rawMarkdown": "Hi, has anyone tried out predicting using ARIMA or LSTM? It seems that people are just using lightgbm with little consideration that it is a continuous time series problem. \nI'm quite new to machine learning and especially time series prediction, so i have a couple (maybe silly) questions regarding this:\n1. If using ARIMA, are you supposed to do rolling forecasting? if so, it seems the amount of training needed is crazy given how many data points there are. Also how are you supposed to do rolling forecasting with test set? based on lightgbm's predictions?\n2. How does ARIMA with external regressors work? and do these regressors have to be time series as well (temperature etc)? \n3. Is LSTM a good choice for this problem? has anyone tried that?\n4. It was mentioned in another discussion that you can do fourier and ARIMA, how does that work? \nsorry for the amateur questions, any advice is greatly appreciated!",
      "votes": null
    },
    {
      "id": "673325",
      "postDate": "11/14/2019 20:27:56",
      "content": "<p>I think it will be quite trouble some to treat it as timeseries as you will have to forecast for 2 years in future and LSTMs are very suitable , it would take a lot of time doing that . Let me know if you can figure out a way to predict multistep forecast at a faster rate</p>",
      "rawMarkdown": "I think it will be quite trouble some to treat it as timeseries as you will have to forecast for 2 years in future and LSTMs are very suitable , it would take a lot of time doing that . Let me know if you can figure out a way to predict multistep forecast at a faster rate",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 673325,
      "author_name": "avikrams",
      "author_url": "",
      "post_date": "11/14/2019 20:27:56",
      "content": "<p>I think it will be quite trouble some to treat it as timeseries as you will have to forecast for 2 years in future and LSTMs are very suitable , it would take a lot of time doing that . Let me know if you can figure out a way to predict multistep forecast at a faster rate</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "673122": "Hi, has anyone tried out predicting using ARIMA or LSTM? It seems that people are just using lightgbm with little consideration that it is a continuous time series problem. \nI'm quite new to machine learning and especially time series prediction, so i have a couple (maybe silly) questions regarding this:\n1. If using ARIMA, are you supposed to do rolling forecasting? if so, it seems the amount of training needed is crazy given how many data points there are. Also how are you supposed to do rolling forecasting with test set? based on lightgbm's predictions?\n2. How does ARIMA with external regressors work? and do these regressors have to be time series as well (temperature etc)? \n3. Is LSTM a good choice for this problem? has anyone tried that?\n4. It was mentioned in another discussion that you can do fourier and ARIMA, how does that work? \nsorry for the amateur questions, any advice is greatly appreciated!",
    "673325": "I think it will be quite trouble some to treat it as timeseries as you will have to forecast for 2 years in future and LSTMs are very suitable , it would take a lot of time doing that . Let me know if you can figure out a way to predict multistep forecast at a faster rate"
  },
  "source": "meta"
}