{
  "id": 236164,
  "title": "End to End wifi Features model",
  "url": "/competitions/indoor-location-navigation/discussion/236164",
  "author_name": "SuryaJR_Rafl",
  "post_date": "2021-05-03T06:10:27.535000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi,</p>\n<p>I know this is pretty late in the competition, but I just started trying out some ideas I had in mind. I have shared a <a href=\"https://www.kaggle.com/suryajrrafl/end-to-end-wifi-features-model/notebook\" target=\"_blank\">notebook</a>  where I have tried to address pytorch models using WiFi features data end-to-end. </p>\n<ol>\n<li>setting up a config class to tune parameters easily</li>\n<li>Deterministic seed function for torch users</li>\n<li>LR range finder function</li>\n<li>This includes reading data, creating Pytorch Datasets and Dataloaders</li>\n<li>Pytorch model to fit (in this case a simple MLP model)</li>\n<li>Choosing cv strategy (groupkfold seems correct for the competition)</li>\n<li>train models according to building wifi features Data</li>\n<li>Generate OOF predictions from each fold's validation data</li>\n<li>Predicting for Test set</li>\n<li>plotting training results</li>\n</ol>\n<p>If you're new to OOF concept, try reading <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175614\" target=\"_blank\">chris doette's wonderful post on hill climbing method</a></p>\n<p>As of now I haven't added any post processing, but will try to add it here if time permits. Feedback is most welcome. </p>",
  "messages": [
    {
      "id": 1291564,
      "postDate": "2021-05-03T06:10:27.537Z",
      "content": "<p>Hi,</p>\n<p>I know this is pretty late in the competition, but I just started trying out some ideas I had in mind. I have shared a <a href=\"https://www.kaggle.com/suryajrrafl/end-to-end-wifi-features-model/notebook\" target=\"_blank\">notebook</a>  where I have tried to address pytorch models using WiFi features data end-to-end. </p>\n<ol>\n<li>setting up a config class to tune parameters easily</li>\n<li>Deterministic seed function for torch users</li>\n<li>LR range finder function</li>\n<li>This includes reading data, creating Pytorch Datasets and Dataloaders</li>\n<li>Pytorch model to fit (in this case a simple MLP model)</li>\n<li>Choosing cv strategy (groupkfold seems correct for the competition)</li>\n<li>train models according to building wifi features Data</li>\n<li>Generate OOF predictions from each fold's validation data</li>\n<li>Predicting for Test set</li>\n<li>plotting training results</li>\n</ol>\n<p>If you're new to OOF concept, try reading <a href=\"https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175614\" target=\"_blank\">chris doette's wonderful post on hill climbing method</a></p>\n<p>As of now I haven't added any post processing, but will try to add it here if time permits. Feedback is most welcome. </p>",
      "rawMarkdown": "Hi,\n\nI know this is pretty late in the competition, but I just started trying out some ideas I had in mind. I have shared a [notebook](https://www.kaggle.com/suryajrrafl/end-to-end-wifi-features-model/notebook)  where I have tried to address pytorch models using WiFi features data end-to-end. \n\n\n1. setting up a config class to tune parameters easily\n2. Deterministic seed function for torch users\n3. LR range finder function\n4. This includes reading data, creating Pytorch Datasets and Dataloaders\n5. Pytorch model to fit (in this case a simple MLP model)\n6. Choosing cv strategy (groupkfold seems correct for the competition)\n7. train models according to building wifi features Data\n8. Generate OOF predictions from each fold's validation data\n9. Predicting for Test set\n10. plotting training results\n\nIf you're new to OOF concept, try reading [chris doette's wonderful post on hill climbing method](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175614)\n\nAs of now I haven't added any post processing, but will try to add it here if time permits. Feedback is most welcome. ",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1291564": "Hi,\n\nI know this is pretty late in the competition, but I just started trying out some ideas I had in mind. I have shared a [notebook](https://www.kaggle.com/suryajrrafl/end-to-end-wifi-features-model/notebook)  where I have tried to address pytorch models using WiFi features data end-to-end. \n\n\n1. setting up a config class to tune parameters easily\n2. Deterministic seed function for torch users\n3. LR range finder function\n4. This includes reading data, creating Pytorch Datasets and Dataloaders\n5. Pytorch model to fit (in this case a simple MLP model)\n6. Choosing cv strategy (groupkfold seems correct for the competition)\n7. train models according to building wifi features Data\n8. Generate OOF predictions from each fold's validation data\n9. Predicting for Test set\n10. plotting training results\n\nIf you're new to OOF concept, try reading [chris doette's wonderful post on hill climbing method](https://www.kaggle.com/c/siim-isic-melanoma-classification/discussion/175614)\n\nAs of now I haven't added any post processing, but will try to add it here if time permits. Feedback is most welcome. "
  }
}