{
  "id": 240056,
  "title": "Part of the 30th place solution",
  "url": "/competitions/indoor-location-navigation/writeups/may-the-shakeup-be-with-you-part-of-the-30th-place",
  "author_name": "",
  "post_date": "2021-05-18T12:26:34.241237200Z",
  "votes": 22,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Congratulations to all the winners! <br>\nIt was actually a tough competition but we enjoyed and learned a lot!<br>\nI want to say thank you to all the kagglers who participated in the comp! </p>\n<p>Our team got 30th place, and actually my model was not so important for our result.<br>\nTeam mates' models perform far better than mine.<br>\nBut our team agreed that we won't publish our solution. <br>\nStill, I think I want (or need) to publish some of my work so I posted it. </p>\n<h2>The LSMT notebook</h2>\n<p>I published <a href=\"https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats\" target=\"_blank\">the LSTM notebook</a> 2 months ago, and I want to show how it went after that. </p>\n<p><a href=\"https://www.kaggle.com/kokitanisaka/self-attentintive-lstm-by-keras\" target=\"_blank\">This is the notebook. </a></p>\n<p>I applied self-attention layer and some more modifications so it got better. <br>\nActually the performance of the notebook is so so. Public score is 6.060. <br>\nFeel free to throw any comments or feedback. Thanks!</p>\n<h2>Post process</h2>\n<p>I tried a post process but it didn't work for public LB so we didn't use it.<br>\nBut when I see the private score, it seems it worked. <br>\nSo I choose to publish the following notebook as well.</p>\n<p>The idea is to fix the result of the snap to grid. <br>\nWe thought that the ground truth should be on grids, so we relied on snap to grid.<br>\nBut after we applied snap to grid, some paths looked definitely wrong, so I tackled the issue. </p>\n<p><a href=\"https://www.kaggle.com/kokitanisaka/create-arrayed-map\" target=\"_blank\">This is the preparation for the pp.</a><br>\n<a href=\"https://www.kaggle.com/kokitanisaka/fix-snapped-waypoints\" target=\"_blank\">And this is the pp.</a></p>\n<p>Again, thank you to all of you! I got another memorable competition that I joined. I really enjoyed it! <br>\nLook forward to seeing you in other competitions! </p>",
  "messages": [
    {
      "id": "1313149",
      "postDate": "05/18/2021 12:26:34",
      "content": "<p>Congratulations to all the winners! <br>\nIt was actually a tough competition but we enjoyed and learned a lot!<br>\nI want to say thank you to all the kagglers who participated in the comp! </p>\n<p>Our team got 30th place, and actually my model was not so important for our result.<br>\nTeam mates' models perform far better than mine.<br>\nBut our team agreed that we won't publish our solution. <br>\nStill, I think I want (or need) to publish some of my work so I posted it. </p>\n<h2>The LSMT notebook</h2>\n<p>I published <a href=\"https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats\" target=\"_blank\">the LSTM notebook</a> 2 months ago, and I want to show how it went after that. </p>\n<p><a href=\"https://www.kaggle.com/kokitanisaka/self-attentintive-lstm-by-keras\" target=\"_blank\">This is the notebook. </a></p>\n<p>I applied self-attention layer and some more modifications so it got better. <br>\nActually the performance of the notebook is so so. Public score is 6.060. <br>\nFeel free to throw any comments or feedback. Thanks!</p>\n<h2>Post process</h2>\n<p>I tried a post process but it didn't work for public LB so we didn't use it.<br>\nBut when I see the private score, it seems it worked. <br>\nSo I choose to publish the following notebook as well.</p>\n<p>The idea is to fix the result of the snap to grid. <br>\nWe thought that the ground truth should be on grids, so we relied on snap to grid.<br>\nBut after we applied snap to grid, some paths looked definitely wrong, so I tackled the issue. </p>\n<p><a href=\"https://www.kaggle.com/kokitanisaka/create-arrayed-map\" target=\"_blank\">This is the preparation for the pp.</a><br>\n<a href=\"https://www.kaggle.com/kokitanisaka/fix-snapped-waypoints\" target=\"_blank\">And this is the pp.</a></p>\n<p>Again, thank you to all of you! I got another memorable competition that I joined. I really enjoyed it! <br>\nLook forward to seeing you in other competitions! </p>",
      "rawMarkdown": "Congratulations to all the winners! \nIt was actually a tough competition but we enjoyed and learned a lot!\nI want to say thank you to all the kagglers who participated in the comp! \n\nOur team got 30th place, and actually my model was not so important for our result.\nTeam mates' models perform far better than mine.\nBut our team agreed that we won't publish our solution. \nStill, I think I want (or need) to publish some of my work so I posted it. \n\n## The LSMT notebook\nI published [the LSTM notebook](https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats) 2 months ago, and I want to show how it went after that. \n\n[This is the notebook. ](https://www.kaggle.com/kokitanisaka/self-attentintive-lstm-by-keras)\n\nI applied self-attention layer and some more modifications so it got better. \nActually the performance of the notebook is so so. Public score is 6.060. \nFeel free to throw any comments or feedback. Thanks!\n\n## Post process\nI tried a post process but it didn't work for public LB so we didn't use it.\nBut when I see the private score, it seems it worked. \nSo I choose to publish the following notebook as well.\n\nThe idea is to fix the result of the snap to grid. \nWe thought that the ground truth should be on grids, so we relied on snap to grid.\nBut after we applied snap to grid, some paths looked definitely wrong, so I tackled the issue. \n\n[This is the preparation for the pp.](https://www.kaggle.com/kokitanisaka/create-arrayed-map)\n[And this is the pp.](https://www.kaggle.com/kokitanisaka/fix-snapped-waypoints)\n\nAgain, thank you to all of you! I got another memorable competition that I joined. I really enjoyed it! \nLook forward to seeing you in other competitions!",
      "votes": null
    },
    {
      "id": "1313183",
      "postDate": "05/18/2021 12:38:32",
      "content": "<p>Thanks for sharing your writeup and congratulations on becoming competitions master.Your lstm baseline was very helpful in this comeptition.</p>",
      "rawMarkdown": "Thanks for sharing your writeup and congratulations on becoming competitions master.Your lstm baseline was very helpful in this comeptition.",
      "votes": null
    },
    {
      "id": "1314030",
      "postDate": "05/18/2021 22:19:04",
      "content": "<p>Thank you for the kind comment! </p>",
      "rawMarkdown": "Thank you for the kind comment!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1313183,
      "author_name": "nischaydnk",
      "author_url": "",
      "post_date": "05/18/2021 12:38:32",
      "content": "<p>Thanks for sharing your writeup and congratulations on becoming competitions master.Your lstm baseline was very helpful in this comeptition.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1314030,
          "author_name": "kokitanisaka",
          "author_url": "",
          "post_date": "05/18/2021 22:19:04",
          "content": "<p>Thank you for the kind comment! </p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1313149": "Congratulations to all the winners! \nIt was actually a tough competition but we enjoyed and learned a lot!\nI want to say thank you to all the kagglers who participated in the comp! \n\nOur team got 30th place, and actually my model was not so important for our result.\nTeam mates' models perform far better than mine.\nBut our team agreed that we won't publish our solution. \nStill, I think I want (or need) to publish some of my work so I posted it. \n\n## The LSMT notebook\nI published [the LSTM notebook](https://www.kaggle.com/kokitanisaka/lstm-by-keras-with-unified-wi-fi-feats) 2 months ago, and I want to show how it went after that. \n\n[This is the notebook. ](https://www.kaggle.com/kokitanisaka/self-attentintive-lstm-by-keras)\n\nI applied self-attention layer and some more modifications so it got better. \nActually the performance of the notebook is so so. Public score is 6.060. \nFeel free to throw any comments or feedback. Thanks!\n\n## Post process\nI tried a post process but it didn't work for public LB so we didn't use it.\nBut when I see the private score, it seems it worked. \nSo I choose to publish the following notebook as well.\n\nThe idea is to fix the result of the snap to grid. \nWe thought that the ground truth should be on grids, so we relied on snap to grid.\nBut after we applied snap to grid, some paths looked definitely wrong, so I tackled the issue. \n\n[This is the preparation for the pp.](https://www.kaggle.com/kokitanisaka/create-arrayed-map)\n[And this is the pp.](https://www.kaggle.com/kokitanisaka/fix-snapped-waypoints)\n\nAgain, thank you to all of you! I got another memorable competition that I joined. I really enjoyed it! \nLook forward to seeing you in other competitions!",
    "1313183": "Thanks for sharing your writeup and congratulations on becoming competitions master.Your lstm baseline was very helpful in this comeptition.",
    "1314030": "Thank you for the kind comment!"
  },
  "source": "meta"
}