{
  "id": 242289,
  "title": "post-processing by outlier correction",
  "url": "/competitions/google-smartphone-decimeter-challenge/discussion/242289",
  "author_name": "dehokanta",
  "post_date": "2021-05-28T11:31:45.351000",
  "votes": 11,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I have published a simple post-processing notebook with outlier correction.<br>\nWhat do you think about this process? I look forward to your comments.</p>\n<p><a href=\"https://www.kaggle.com/dehokanta/baseline-post-processing-by-outlier-correction\" target=\"_blank\">https://www.kaggle.com/dehokanta/baseline-post-processing-by-outlier-correction</a></p>",
  "messages": [
    {
      "id": 1326295,
      "postDate": "2021-05-28T11:31:45.350Z",
      "content": "<p>I have published a simple post-processing notebook with outlier correction.<br>\nWhat do you think about this process? I look forward to your comments.</p>\n<p><a href=\"https://www.kaggle.com/dehokanta/baseline-post-processing-by-outlier-correction\" target=\"_blank\">https://www.kaggle.com/dehokanta/baseline-post-processing-by-outlier-correction</a></p>",
      "rawMarkdown": "I have published a simple post-processing notebook with outlier correction.\nWhat do you think about this process? I look forward to your comments.\n\nhttps://www.kaggle.com/dehokanta/baseline-post-processing-by-outlier-correction",
      "votes": 11
    },
    {
      "id": 1328981,
      "postDate": "2021-05-30T18:06:21.613Z",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/dehokanta\" target=\"_blank\">@dehokanta</a> <br>\nGood ideas so far. Congratulations !<br>\nCan you please elaborate for a non expert in Kalman filters how did you choose the state transitions matrix and what are the 6 parameters?<br>\nstate_transition = np.array([[1, 0, T, 0, 0.5 * T ** 2, 0], [0, 1, 0, T, 0, 0.5 * T ** 2], [0, 0, 1, 0, T, 0],<br>\n                             [0, 0, 0, 1, 0, T], [0, 0, 0, 0, 1, 0], [0, 0, 0, 0, 0, 1]])</p>",
      "rawMarkdown": "Hi @dehokanta \nGood ideas so far. Congratulations !\nCan you please elaborate for a non expert in Kalman filters how did you choose the state transitions matrix and what are the 6 parameters?\nstate_transition = np.array([[1, 0, T, 0, 0.5 * T ** 2, 0], [0, 1, 0, T, 0, 0.5 * T ** 2], [0, 0, 1, 0, T, 0],\n                             [0, 0, 0, 1, 0, T], [0, 0, 0, 0, 1, 0], [0, 0, 0, 0, 0, 1]])\n",
      "votes": 1,
      "replies": [
        {
          "id": 1329078,
          "postDate": "2021-05-30T20:07:50.400Z",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/vladvdv\" target=\"_blank\">@vladvdv</a> <br>\nThanks for asking.Unfortunately, I'm not an expert on Kalman filters too.</p>\n<p>The code was referenced from this notebook.<br>\n<a href=\"https://www.kaggle.com/emaerthin/demonstration-of-the-kalman-filter/comments\" target=\"_blank\">https://www.kaggle.com/emaerthin/demonstration-of-the-kalman-filter/comments</a></p>\n<p>You may get a better answer by asking Marcin Bodych.<br>\nI'm sorry I can't help you.</p>",
          "rawMarkdown": "Hi @vladvdv \nThanks for asking.Unfortunately, I'm not an expert on Kalman filters too.\n\nThe code was referenced from this notebook.\nhttps://www.kaggle.com/emaerthin/demonstration-of-the-kalman-filter/comments\n\nYou may get a better answer by asking Marcin Bodych.\nI'm sorry I can't help you."
        },
        {
          "id": 1329158,
          "postDate": "2021-05-30T22:42:46.087Z",
          "rawMarkdown": "",
          "votes": 4,
          "isDeleted": true
        },
        {
          "id": 1329277,
          "postDate": "2021-05-31T03:07:41.400Z",
          "content": "<p>Thanks for following <a href=\"https://www.kaggle.com/andryrafaralahy\" target=\"_blank\">@andryrafaralahy</a> </p>",
          "rawMarkdown": "Thanks for following @andryrafaralahy "
        },
        {
          "id": 1329535,
          "postDate": "2021-05-31T07:52:00.937Z",
          "content": "<p>Thank you <a href=\"https://www.kaggle.com/andryrafaralahy\" target=\"_blank\">@andryrafaralahy</a> It makes sense <br>\nSo in the Kalman filter you designed, the GPS training data are just used to check the accuracy of the parameters after you smooth (with the calc_haversine). There is no training done per se. And then you smooth the test data with the Kalman model. Practically, the Kalman filter is used as a smoothing method in this case. <br>\nAlso, from my understanding the only adjustable parameters are the 2 noises . The time is 1 sec (there are 1 seconds between the measurements) and the equations are the classical equations from physics.</p>",
          "rawMarkdown": "Thank you @andryrafaralahy It makes sense \nSo in the Kalman filter you designed, the GPS training data are just used to check the accuracy of the parameters after you smooth (with the calc_haversine). There is no training done per se. And then you smooth the test data with the Kalman model. Practically, the Kalman filter is used as a smoothing method in this case. \nAlso, from my understanding the only adjustable parameters are the 2 noises . The time is 1 sec (there are 1 seconds between the measurements) and the equations are the classical equations from physics.\n"
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 1328981,
      "author_name": "Vlad Vaduva",
      "author_url": "",
      "post_date": "2021-05-30T18:06:21.613000",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/dehokanta\" target=\"_blank\">@dehokanta</a> <br>\nGood ideas so far. Congratulations !<br>\nCan you please elaborate for a non expert in Kalman filters how did you choose the state transitions matrix and what are the 6 parameters?<br>\nstate_transition = np.array([[1, 0, T, 0, 0.5 * T ** 2, 0], [0, 1, 0, T, 0, 0.5 * T ** 2], [0, 0, 1, 0, T, 0],<br>\n                             [0, 0, 0, 1, 0, T], [0, 0, 0, 0, 1, 0], [0, 0, 0, 0, 0, 1]])</p>",
      "votes": 1,
      "replies": [
        {
          "id": 1329078,
          "author_name": "dehokanta",
          "author_url": "",
          "post_date": "2021-05-30T20:07:50.400000",
          "content": "<p>Hi <a href=\"https://www.kaggle.com/vladvdv\" target=\"_blank\">@vladvdv</a> <br>\nThanks for asking.Unfortunately, I'm not an expert on Kalman filters too.</p>\n<p>The code was referenced from this notebook.<br>\n<a href=\"https://www.kaggle.com/emaerthin/demonstration-of-the-kalman-filter/comments\" target=\"_blank\">https://www.kaggle.com/emaerthin/demonstration-of-the-kalman-filter/comments</a></p>\n<p>You may get a better answer by asking Marcin Bodych.<br>\nI'm sorry I can't help you.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1329158,
          "author_name": "",
          "author_url": "",
          "post_date": "2021-05-30T22:42:46.087000",
          "content": "",
          "votes": 4,
          "replies": []
        },
        {
          "id": 1329277,
          "author_name": "dehokanta",
          "author_url": "",
          "post_date": "2021-05-31T03:07:41.400000",
          "content": "<p>Thanks for following <a href=\"https://www.kaggle.com/andryrafaralahy\" target=\"_blank\">@andryrafaralahy</a> </p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 1329535,
          "author_name": "Vlad Vaduva",
          "author_url": "",
          "post_date": "2021-05-31T07:52:00.937000",
          "content": "<p>Thank you <a href=\"https://www.kaggle.com/andryrafaralahy\" target=\"_blank\">@andryrafaralahy</a> It makes sense <br>\nSo in the Kalman filter you designed, the GPS training data are just used to check the accuracy of the parameters after you smooth (with the calc_haversine). There is no training done per se. And then you smooth the test data with the Kalman model. Practically, the Kalman filter is used as a smoothing method in this case. <br>\nAlso, from my understanding the only adjustable parameters are the 2 noises . The time is 1 sec (there are 1 seconds between the measurements) and the equations are the classical equations from physics.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1326295": "I have published a simple post-processing notebook with outlier correction.\nWhat do you think about this process? I look forward to your comments.\n\nhttps://www.kaggle.com/dehokanta/baseline-post-processing-by-outlier-correction",
    "1328981": "Hi @dehokanta \nGood ideas so far. Congratulations !\nCan you please elaborate for a non expert in Kalman filters how did you choose the state transitions matrix and what are the 6 parameters?\nstate_transition = np.array([[1, 0, T, 0, 0.5 * T ** 2, 0], [0, 1, 0, T, 0, 0.5 * T ** 2], [0, 0, 1, 0, T, 0],\n                             [0, 0, 0, 1, 0, T], [0, 0, 0, 0, 1, 0], [0, 0, 0, 0, 0, 1]])\n"
  }
}