{
  "id": 230096,
  "title": "Effect of the leak?",
  "url": "/competitions/indoor-location-navigation/discussion/230096",
  "author_name": "",
  "post_date": "2021-04-01T23:44:06.994581500Z",
  "votes": 13,
  "comment_count": 6,
  "views": 0,
  "content": "<p>I was really surprised at <a href=\"https://www.kaggle.com/tomooinubushi/postprocessing-based-on-leakage\" target=\"_blank\">https://www.kaggle.com/tomooinubushi/postprocessing-based-on-leakage</a> and I applied the postprocessing to my submission and this is the result. My model (3.091) doesn't use raw timestamp as input and the post processing result is 3.055, so the improvement is about 0.04, which looks small for me. However, I'm really afraid that there maybe exists a better way to leverage the leak information…<br>\n<a href=\"https://imgur.com/78wnrrz\" target=\"_blank\">https://imgur.com/78wnrrz</a></p>",
  "messages": [
    {
      "id": "1260207",
      "postDate": "04/01/2021 23:44:06",
      "content": "<p>I was really surprised at <a href=\"https://www.kaggle.com/tomooinubushi/postprocessing-based-on-leakage\" target=\"_blank\">https://www.kaggle.com/tomooinubushi/postprocessing-based-on-leakage</a> and I applied the postprocessing to my submission and this is the result. My model (3.091) doesn't use raw timestamp as input and the post processing result is 3.055, so the improvement is about 0.04, which looks small for me. However, I'm really afraid that there maybe exists a better way to leverage the leak information…<br>\n<a href=\"https://imgur.com/78wnrrz\" target=\"_blank\">https://imgur.com/78wnrrz</a></p>",
      "rawMarkdown": "I was really surprised at https://www.kaggle.com/tomooinubushi/postprocessing-based-on-leakage and I applied the postprocessing to my submission and this is the result. My model (3.091) doesn't use raw timestamp as input and the post processing result is 3.055, so the improvement is about 0.04, which looks small for me. However, I'm really afraid that there maybe exists a better way to leverage the leak information...\n[https://imgur.com/78wnrrz](https://imgur.com/78wnrrz)",
      "votes": null
    },
    {
      "id": "1260225",
      "postDate": "04/02/2021 00:14:51",
      "content": "<p>Thanks for sharing the result. I was going to do something similar. Btw now we have some little clue about your awesome model :)</p>",
      "rawMarkdown": "Thanks for sharing the result. I was going to do something similar. Btw now we have some little clue about your awesome model :)",
      "votes": null
    },
    {
      "id": "1260272",
      "postDate": "04/02/2021 01:44:47",
      "content": "<p>Thank you for sharing your results. I was interested in whether this simple postprocessing method could be applicable for better results. <br>\nIf you do not change my time threshold (5,500) and postprocess both start and endpoints. You may postprocess 134 samples of 10,133 waypoints (1.3%) and got 0.036 gain. The gain itself is small, but I could not estimate the potential of the leakage.</p>",
      "rawMarkdown": "Thank you for sharing your results. I was interested in whether this simple postprocessing method could be applicable for better results. \nIf you do not change my time threshold (5,500) and postprocess both start and endpoints. You may postprocess 134 samples of 10,133 waypoints (1.3%) and got 0.036 gain. The gain itself is small, but I could not estimate the potential of the leakage.",
      "votes": null
    },
    {
      "id": "1260276",
      "postDate": "04/02/2021 01:53:30",
      "content": "<p>I didn't change anything from your kernel and use both start and end points now. yeah, I need to check the potential of the leakage.</p>",
      "rawMarkdown": "I didn't change anything from your kernel and use both start and end points now. yeah, I need to check the potential of the leakage.",
      "votes": null
    },
    {
      "id": "1262088",
      "postDate": "04/03/2021 19:05:25",
      "content": "<p>thanks for sharing </p>",
      "rawMarkdown": "thanks for sharing",
      "votes": null
    },
    {
      "id": "1262373",
      "postDate": "04/04/2021 07:21:54",
      "content": "<p>I tried to use leak information as features yesterday, but not succeeded yet. seems it's not so easy to leverage it.</p>",
      "rawMarkdown": "I tried to use leak information as features yesterday, but not succeeded yet. seems it's not so easy to leverage it.",
      "votes": null
    },
    {
      "id": "1264132",
      "postDate": "04/05/2021 23:02:02",
      "content": "<p><a href=\"https://www.kaggle.com/mamasinkgs\" target=\"_blank\">@mamasinkgs</a> I'm just blown away you are almost sub 3 and I cannot get sub 4 with my score 😂<br>\nI haven't tried anything with the leak. I got a long way to go just working my features to even try to jump anywhere near gold. 🙏</p>",
      "rawMarkdown": "mamasinkgs I'm just blown away you are almost sub 3 and I cannot get sub 4 with my score 😂\nI haven't tried anything with the leak. I got a long way to go just working my features to even try to jump anywhere near gold. 🙏",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1260225,
      "author_name": "higepon",
      "author_url": "",
      "post_date": "04/02/2021 00:14:51",
      "content": "<p>Thanks for sharing the result. I was going to do something similar. Btw now we have some little clue about your awesome model :)</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1260272,
      "author_name": "tomooinubushi",
      "author_url": "",
      "post_date": "04/02/2021 01:44:47",
      "content": "<p>Thank you for sharing your results. I was interested in whether this simple postprocessing method could be applicable for better results. <br>\nIf you do not change my time threshold (5,500) and postprocess both start and endpoints. You may postprocess 134 samples of 10,133 waypoints (1.3%) and got 0.036 gain. The gain itself is small, but I could not estimate the potential of the leakage.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1260276,
          "author_name": "mamasinkgs",
          "author_url": "",
          "post_date": "04/02/2021 01:53:30",
          "content": "<p>I didn't change anything from your kernel and use both start and end points now. yeah, I need to check the potential of the leakage.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1262373,
          "author_name": "mamasinkgs",
          "author_url": "",
          "post_date": "04/04/2021 07:21:54",
          "content": "<p>I tried to use leak information as features yesterday, but not succeeded yet. seems it's not so easy to leverage it.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1262088,
      "author_name": "mohamedbakrey",
      "author_url": "",
      "post_date": "04/03/2021 19:05:25",
      "content": "<p>thanks for sharing </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1264132,
      "author_name": "crained",
      "author_url": "",
      "post_date": "04/05/2021 23:02:02",
      "content": "<p><a href=\"https://www.kaggle.com/mamasinkgs\" target=\"_blank\">@mamasinkgs</a> I'm just blown away you are almost sub 3 and I cannot get sub 4 with my score 😂<br>\nI haven't tried anything with the leak. I got a long way to go just working my features to even try to jump anywhere near gold. 🙏</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1260207": "I was really surprised at https://www.kaggle.com/tomooinubushi/postprocessing-based-on-leakage and I applied the postprocessing to my submission and this is the result. My model (3.091) doesn't use raw timestamp as input and the post processing result is 3.055, so the improvement is about 0.04, which looks small for me. However, I'm really afraid that there maybe exists a better way to leverage the leak information...\n[https://imgur.com/78wnrrz](https://imgur.com/78wnrrz)",
    "1260225": "Thanks for sharing the result. I was going to do something similar. Btw now we have some little clue about your awesome model :)",
    "1260272": "Thank you for sharing your results. I was interested in whether this simple postprocessing method could be applicable for better results. \nIf you do not change my time threshold (5,500) and postprocess both start and endpoints. You may postprocess 134 samples of 10,133 waypoints (1.3%) and got 0.036 gain. The gain itself is small, but I could not estimate the potential of the leakage.",
    "1260276": "I didn't change anything from your kernel and use both start and end points now. yeah, I need to check the potential of the leakage.",
    "1262088": "thanks for sharing",
    "1262373": "I tried to use leak information as features yesterday, but not succeeded yet. seems it's not so easy to leverage it.",
    "1264132": "mamasinkgs I'm just blown away you are almost sub 3 and I cannot get sub 4 with my score 😂\nI haven't tried anything with the leak. I got a long way to go just working my features to even try to jump anywhere near gold. 🙏"
  },
  "source": "meta"
}