{
  "id": 223834,
  "title": "Dealing with Wi-Fi features with LSTM using all the sites",
  "url": "/competitions/indoor-location-navigation/discussion/223834",
  "author_name": "",
  "post_date": "2021-03-05T21:00:27.754475400Z",
  "votes": 16,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I published <a href=\"https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats/execution\" target=\"_blank\">a notebook</a>, how we can utilize <a href=\"https://www.kaggle.com/kokitanisaka/indoorunifiedwifids\" target=\"_blank\">the unified wifi dataset</a> with RNN.</p>\n<p>I felt that some people are suffered from how to use the dataset, <br>\nthen I thought I should publish an example.</p>\n<p>This is not the best scored notebook of mine,<br>\nbut you will see the way I incorporate these features into RNN.</p>\n<p>I just made it work, so I think there are many other ways to utilize the dataset.<br>\nAnd also there should be much space to improve its accuracy.</p>\n<p>I hope it can help, thanks!</p>",
  "messages": [
    {
      "id": "1227776",
      "postDate": "03/05/2021 21:00:27",
      "content": "<p>I published <a href=\"https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats/execution\" target=\"_blank\">a notebook</a>, how we can utilize <a href=\"https://www.kaggle.com/kokitanisaka/indoorunifiedwifids\" target=\"_blank\">the unified wifi dataset</a> with RNN.</p>\n<p>I felt that some people are suffered from how to use the dataset, <br>\nthen I thought I should publish an example.</p>\n<p>This is not the best scored notebook of mine,<br>\nbut you will see the way I incorporate these features into RNN.</p>\n<p>I just made it work, so I think there are many other ways to utilize the dataset.<br>\nAnd also there should be much space to improve its accuracy.</p>\n<p>I hope it can help, thanks!</p>",
      "rawMarkdown": "I published [a notebook](https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats/execution), how we can utilize [the unified wifi dataset](https://www.kaggle.com/kokitanisaka/indoorunifiedwifids) with RNN.\n\nI felt that some people are suffered from how to use the dataset, \nthen I thought I should publish an example.\n\nThis is not the best scored notebook of mine,\nbut you will see the way I incorporate these features into RNN.\n\nI just made it work, so I think there are many other ways to utilize the dataset.\nAnd also there should be much space to improve its accuracy.\n\nI hope it can help, thanks!",
      "votes": null
    },
    {
      "id": "1230059",
      "postDate": "03/07/2021 18:44:42",
      "content": "<p>Hi Kouki, thank you for sharing this. Would you mind sharing your CV score for this approach?</p>",
      "rawMarkdown": "Hi Kouki, thank you for sharing this. Would you mind sharing your CV score for this approach?",
      "votes": null
    },
    {
      "id": "1230207",
      "postDate": "03/07/2021 22:10:02",
      "content": "<p>Hi hrunic! <br>\nThat's the thing. As it takes much time to finish training, I haven't gotten a proper CV score yet.<br>\nMy current model takes half a day for one fold on my own PC (I don't have GPU). <br>\nFor now, what I'm doing is tune the model seeing the validation loss. </p>",
      "rawMarkdown": "Hi hrunic! \nThat's the thing. As it takes much time to finish training, I haven't gotten a proper CV score yet.\nMy current model takes half a day for one fold on my own PC (I don't have GPU). \nFor now, what I'm doing is tune the model seeing the validation loss.",
      "votes": null
    },
    {
      "id": "1230460",
      "postDate": "03/08/2021 07:10:48",
      "content": "<p><a href=\"https://www.kaggle.com/kokitanisaka\" target=\"_blank\">@kokitanisaka</a> Oh, half a day is tedious. For me locally its about 40 secs per epoch with a similar approach. Why not use Kaggle Notebooks? Anyways, thanks for the info!</p>",
      "rawMarkdown": "kokitanisaka Oh, half a day is tedious. For me locally its about 40 secs per epoch with a similar approach. Why not use Kaggle Notebooks? Anyways, thanks for the info!",
      "votes": null
    },
    {
      "id": "1230505",
      "postDate": "03/08/2021 07:44:26",
      "content": "<p>When I train the model with my local PC's CPU, it's faster and still it takes half a day.<br>\nIf I use kaggle notebook, it won't finish in 9 hours so I can't have the trained model that's why. 😅<br>\nI should make the model or the way I use the data better.</p>",
      "rawMarkdown": "When I train the model with my local PC's CPU, it's faster and still it takes half a day.\nIf I use kaggle notebook, it won't finish in 9 hours so I can't have the trained model that's why. 😅\nI should make the model or the way I use the data better.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1230059,
      "author_name": "nicohrubec",
      "author_url": "",
      "post_date": "03/07/2021 18:44:42",
      "content": "<p>Hi Kouki, thank you for sharing this. Would you mind sharing your CV score for this approach?</p>",
      "votes": null,
      "replies": [
        {
          "id": 1230207,
          "author_name": "kokitanisaka",
          "author_url": "",
          "post_date": "03/07/2021 22:10:02",
          "content": "<p>Hi hrunic! <br>\nThat's the thing. As it takes much time to finish training, I haven't gotten a proper CV score yet.<br>\nMy current model takes half a day for one fold on my own PC (I don't have GPU). <br>\nFor now, what I'm doing is tune the model seeing the validation loss. </p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1230460,
          "author_name": "nicohrubec",
          "author_url": "",
          "post_date": "03/08/2021 07:10:48",
          "content": "<p><a href=\"https://www.kaggle.com/kokitanisaka\" target=\"_blank\">@kokitanisaka</a> Oh, half a day is tedious. For me locally its about 40 secs per epoch with a similar approach. Why not use Kaggle Notebooks? Anyways, thanks for the info!</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1230505,
          "author_name": "kokitanisaka",
          "author_url": "",
          "post_date": "03/08/2021 07:44:26",
          "content": "<p>When I train the model with my local PC's CPU, it's faster and still it takes half a day.<br>\nIf I use kaggle notebook, it won't finish in 9 hours so I can't have the trained model that's why. 😅<br>\nI should make the model or the way I use the data better.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1227776": "I published [a notebook](https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats/execution), how we can utilize [the unified wifi dataset](https://www.kaggle.com/kokitanisaka/indoorunifiedwifids) with RNN.\n\nI felt that some people are suffered from how to use the dataset, \nthen I thought I should publish an example.\n\nThis is not the best scored notebook of mine,\nbut you will see the way I incorporate these features into RNN.\n\nI just made it work, so I think there are many other ways to utilize the dataset.\nAnd also there should be much space to improve its accuracy.\n\nI hope it can help, thanks!",
    "1230059": "Hi Kouki, thank you for sharing this. Would you mind sharing your CV score for this approach?",
    "1230207": "Hi hrunic! \nThat's the thing. As it takes much time to finish training, I haven't gotten a proper CV score yet.\nMy current model takes half a day for one fold on my own PC (I don't have GPU). \nFor now, what I'm doing is tune the model seeing the validation loss.",
    "1230460": "kokitanisaka Oh, half a day is tedious. For me locally its about 40 secs per epoch with a similar approach. Why not use Kaggle Notebooks? Anyways, thanks for the info!",
    "1230505": "When I train the model with my local PC's CPU, it's faster and still it takes half a day.\nIf I use kaggle notebook, it won't finish in 9 hours so I can't have the trained model that's why. 😅\nI should make the model or the way I use the data better."
  },
  "source": "meta"
}