{
  "id": 71228,
  "title": "LSTM for time series analysis",
  "url": "/competitions/PLAsTiCC-2018/discussion/71228",
  "author_name": "",
  "post_date": "2018-11-11T16:51:41.836202100Z",
  "votes": 6,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hi all!</p>\n\n<p>I was wondering if anyone is using an LSTM. We have tried to input the flux and the time difference between the series elements, however we can't get the training loss to decrease very much, and the validation loss is of course even worse. Was anyone able to create a successful LSTM model or is the data not suited for it? Maybe there are not enough examples, or the asynchronous time steps are hard to deal with for the network. Any ideas?</p>\n\n<p>Cheers!</p>",
  "messages": [
    {
      "id": "419296",
      "postDate": "11/11/2018 16:51:41",
      "content": "<p>Hi all!</p>\n\n<p>I was wondering if anyone is using an LSTM. We have tried to input the flux and the time difference between the series elements, however we can't get the training loss to decrease very much, and the validation loss is of course even worse. Was anyone able to create a successful LSTM model or is the data not suited for it? Maybe there are not enough examples, or the asynchronous time steps are hard to deal with for the network. Any ideas?</p>\n\n<p>Cheers!</p>",
      "rawMarkdown": "Hi all!\n\nI was wondering if anyone is using an LSTM. We have tried to input the flux and the time difference between the series elements, however we can't get the training loss to decrease very much, and the validation loss is of course even worse. Was anyone able to create a successful LSTM model or is the data not suited for it? Maybe there are not enough examples, or the asynchronous time steps are hard to deal with for the network. Any ideas?\n\nCheers!",
      "votes": null
    },
    {
      "id": "423360",
      "postDate": "11/18/2018 02:31:33",
      "content": "<p>I am looking to use LSTM also, but running into some challenges getting the data into a format appropriate for LSTM.</p>\n\n<p>What methods did you use to convert the data into sequences for the LSTM? How did you handle the multiple bands and the different timestamps?</p>",
      "rawMarkdown": "I am looking to use LSTM also, but running into some challenges getting the data into a format appropriate for LSTM.\n\nWhat methods did you use to convert the data into sequences for the LSTM? How did you handle the multiple bands and the different timestamps?",
      "votes": null
    },
    {
      "id": "423605",
      "postDate": "11/18/2018 17:18:37",
      "content": "<p>I shared a published one here: <a href=\"https://www.kaggle.com/c/PLAsTiCC-2018/discussion/71949\">https://www.kaggle.com/c/PLAsTiCC-2018/discussion/71949</a></p>",
      "rawMarkdown": "I shared a published one here: https://www.kaggle.com/c/PLAsTiCC-2018/discussion/71949",
      "votes": null
    },
    {
      "id": "518342",
      "postDate": "04/17/2019 05:29:23",
      "content": "<p>Difference between scaling and transformation for a dataset. \ni want to know the difference between the MinMax transformation and log transformation.</p>\n\n<p>should both be applied and when one should be applied , also what does these transformation do to the dataset? \nim working with time series dataset.</p>",
      "rawMarkdown": "Difference between scaling and transformation for a dataset. \ni want to know the difference between the MinMax transformation and log transformation.\n\nshould both be applied and when one should be applied , also what does these transformation do to the dataset? \nim working with time series dataset.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 423360,
      "author_name": "tgibbons",
      "author_url": "",
      "post_date": "11/18/2018 02:31:33",
      "content": "<p>I am looking to use LSTM also, but running into some challenges getting the data into a format appropriate for LSTM.</p>\n\n<p>What methods did you use to convert the data into sequences for the LSTM? How did you handle the multiple bands and the different timestamps?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 423605,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "11/18/2018 17:18:37",
      "content": "<p>I shared a published one here: <a href=\"https://www.kaggle.com/c/PLAsTiCC-2018/discussion/71949\">https://www.kaggle.com/c/PLAsTiCC-2018/discussion/71949</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 518342,
      "author_name": "kunalbasnet",
      "author_url": "",
      "post_date": "04/17/2019 05:29:23",
      "content": "<p>Difference between scaling and transformation for a dataset. \ni want to know the difference between the MinMax transformation and log transformation.</p>\n\n<p>should both be applied and when one should be applied , also what does these transformation do to the dataset? \nim working with time series dataset.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "419296": "Hi all!\n\nI was wondering if anyone is using an LSTM. We have tried to input the flux and the time difference between the series elements, however we can't get the training loss to decrease very much, and the validation loss is of course even worse. Was anyone able to create a successful LSTM model or is the data not suited for it? Maybe there are not enough examples, or the asynchronous time steps are hard to deal with for the network. Any ideas?\n\nCheers!",
    "423360": "I am looking to use LSTM also, but running into some challenges getting the data into a format appropriate for LSTM.\n\nWhat methods did you use to convert the data into sequences for the LSTM? How did you handle the multiple bands and the different timestamps?",
    "423605": "I shared a published one here: https://www.kaggle.com/c/PLAsTiCC-2018/discussion/71949",
    "518342": "Difference between scaling and transformation for a dataset. \ni want to know the difference between the MinMax transformation and log transformation.\n\nshould both be applied and when one should be applied , also what does these transformation do to the dataset? \nim working with time series dataset."
  },
  "source": "meta"
}