{
  "id": 236198,
  "title": "CNN approach to wifi features modelling",
  "url": "/competitions/indoor-location-navigation/discussion/236198",
  "author_name": "SuryaJR_Rafl",
  "post_date": "2021-05-03T10:03:08.489000",
  "votes": 5,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi, </p>\n<p>The wifi features (one-hot encoded version) used in the competition are generally of very high dimension. CNNs can have shared parameters, invariant to translation, which are useful properties for this usecase. I have tried using CNNs for wifi features modelling in this <a href=\"https://www.kaggle.com/suryajrrafl/cnn-approach-to-wifi-features/notebook?scriptVersionId=61691355\" target=\"_blank\">notebook</a> . This was also mentioned by chris in this <a href=\"https://www.kaggle.com/c/indoor-location-navigation/discussion/236096\" target=\"_blank\">discussion</a> . The notebook is not perfect in any way, but has lot of scope for improvement.</p>\n<p><strong>The conv2d layers are very slow compared to MLP layers. Any suggestion on how to improve the speed and general feedback on the approach is most welcome.</strong></p>\n<p>Happy kaggling :) :) :)</p>",
  "messages": [
    {
      "id": 1291749,
      "postDate": "2021-05-03T10:03:08.490Z",
      "content": "<p>Hi, </p>\n<p>The wifi features (one-hot encoded version) used in the competition are generally of very high dimension. CNNs can have shared parameters, invariant to translation, which are useful properties for this usecase. I have tried using CNNs for wifi features modelling in this <a href=\"https://www.kaggle.com/suryajrrafl/cnn-approach-to-wifi-features/notebook?scriptVersionId=61691355\" target=\"_blank\">notebook</a> . This was also mentioned by chris in this <a href=\"https://www.kaggle.com/c/indoor-location-navigation/discussion/236096\" target=\"_blank\">discussion</a> . The notebook is not perfect in any way, but has lot of scope for improvement.</p>\n<p><strong>The conv2d layers are very slow compared to MLP layers. Any suggestion on how to improve the speed and general feedback on the approach is most welcome.</strong></p>\n<p>Happy kaggling :) :) :)</p>",
      "rawMarkdown": "Hi, \n\nThe wifi features (one-hot encoded version) used in the competition are generally of very high dimension. CNNs can have shared parameters, invariant to translation, which are useful properties for this usecase. I have tried using CNNs for wifi features modelling in this [notebook](https://www.kaggle.com/suryajrrafl/cnn-approach-to-wifi-features/notebook?scriptVersionId=61691355) . This was also mentioned by chris in this [discussion](https://www.kaggle.com/c/indoor-location-navigation/discussion/236096) . The notebook is not perfect in any way, but has lot of scope for improvement.\n\n**The conv2d layers are very slow compared to MLP layers. Any suggestion on how to improve the speed and general feedback on the approach is most welcome.**\n\nHappy kaggling :) :) :)",
      "votes": 5
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1291749": "Hi, \n\nThe wifi features (one-hot encoded version) used in the competition are generally of very high dimension. CNNs can have shared parameters, invariant to translation, which are useful properties for this usecase. I have tried using CNNs for wifi features modelling in this [notebook](https://www.kaggle.com/suryajrrafl/cnn-approach-to-wifi-features/notebook?scriptVersionId=61691355) . This was also mentioned by chris in this [discussion](https://www.kaggle.com/c/indoor-location-navigation/discussion/236096) . The notebook is not perfect in any way, but has lot of scope for improvement.\n\n**The conv2d layers are very slow compared to MLP layers. Any suggestion on how to improve the speed and general feedback on the approach is most welcome.**\n\nHappy kaggling :) :) :)"
  }
}