{
  "id": 29162,
  "title": "relabeling mistakes in training data",
  "url": "/competitions/dogs-vs-cats-redux-kernels-edition/discussion/29162",
  "author_name": "",
  "post_date": "2017-02-23T11:57:49.185284700Z",
  "votes": 13,
  "comment_count": 9,
  "views": 0,
  "content": "<p>Hi,</p>\n\n<p>as already brought up in the previous iteration of this competition some of the training data is not exactly perfect. compare here:\n<a href=\"https://www.kaggle.com/c/dogs-vs-cats/discussion/5892\">https://www.kaggle.com/c/dogs-vs-cats/discussion/5892</a></p>\n\n<p>I discovered some of these mistakes just now as I was looking through my models predictions on my cross validation set to see if I could find some systematic problem that would help me improve and get past @tunguz or @guangsha in the final days...</p>\n\n<p>in the spirit of fairness, here are the 52 relabels of the training data that I will now try to use to improve my model.</p>\n\n<p>download attachment or on github:\n<a href=\"https://gist.github.com/lefant/bb9c304cfcf5da0f9f652570cc54893a\">https://gist.github.com/lefant/bb9c304cfcf5da0f9f652570cc54893a</a></p>\n\n<p>cheers and good luck,</p>\n\n<p>Fabian</p>",
  "messages": [
    {
      "id": "163291",
      "postDate": "02/23/2017 11:57:49",
      "content": "<p>Hi,</p>\n\n<p>as already brought up in the previous iteration of this competition some of the training data is not exactly perfect. compare here:\n<a href=\"https://www.kaggle.com/c/dogs-vs-cats/discussion/5892\">https://www.kaggle.com/c/dogs-vs-cats/discussion/5892</a></p>\n\n<p>I discovered some of these mistakes just now as I was looking through my models predictions on my cross validation set to see if I could find some systematic problem that would help me improve and get past @tunguz or @guangsha in the final days...</p>\n\n<p>in the spirit of fairness, here are the 52 relabels of the training data that I will now try to use to improve my model.</p>\n\n<p>download attachment or on github:\n<a href=\"https://gist.github.com/lefant/bb9c304cfcf5da0f9f652570cc54893a\">https://gist.github.com/lefant/bb9c304cfcf5da0f9f652570cc54893a</a></p>\n\n<p>cheers and good luck,</p>\n\n<p>Fabian</p>",
      "rawMarkdown": "Hi,\n\nas already brought up in the previous iteration of this competition some of the training data is not exactly perfect. compare here:\nhttps://www.kaggle.com/c/dogs-vs-cats/discussion/5892\n\nI discovered some of these mistakes just now as I was looking through my models predictions on my cross validation set to see if I could find some systematic problem that would help me improve and get past @tunguz or @guangsha in the final days...\n\nin the spirit of fairness, here are the 52 relabels of the training data that I will now try to use to improve my model.\n\ndownload attachment or on github:\nhttps://gist.github.com/lefant/bb9c304cfcf5da0f9f652570cc54893a\n\ncheers and good luck,\n\nFabian",
      "votes": null
    },
    {
      "id": "163788",
      "postDate": "02/26/2017 00:01:17",
      "content": "<p>Thanks! But I assume the cartoon cats should be labelled 0.1.</p>",
      "rawMarkdown": "Thanks! But I assume the cartoon cats should be labelled 0.1.",
      "votes": null
    },
    {
      "id": "163957",
      "postDate": "02/27/2017 00:57:32",
      "content": "<p>Thanks.  That's really helpful.  I'd done a rudimentary attempt at this but had missed a number of these.  </p>",
      "rawMarkdown": "Thanks.  That's really helpful.  I'd done a rudimentary attempt at this but had missed a number of these.",
      "votes": null
    },
    {
      "id": "163970",
      "postDate": "02/27/2017 03:18:23",
      "content": "<p>Thanks Fabian! I started looking through the training data, but did not get though them yet... Very helpful!</p>",
      "rawMarkdown": "Thanks Fabian! I started looking through the training data, but did not get though them yet... Very helpful!",
      "votes": null
    },
    {
      "id": "165102",
      "postDate": "03/03/2017 18:03:07",
      "content": "<p>Thanks for pointing it out, that's very fair !</p>",
      "rawMarkdown": "Thanks for pointing it out, that's very fair !",
      "votes": null
    },
    {
      "id": "169518",
      "postDate": "03/21/2017 09:31:27",
      "content": "<p>Thanks - very useful</p>",
      "rawMarkdown": "Thanks - very useful",
      "votes": null
    },
    {
      "id": "182817",
      "postDate": "05/16/2017 01:01:34",
      "content": "<p>Thanks! Can you xplain how to find the wrong images via what method?</p>",
      "rawMarkdown": "Thanks! Can you xplain how to find the wrong images via what method?",
      "votes": null
    },
    {
      "id": "300914",
      "postDate": "03/22/2018 03:56:12",
      "content": "<p>Thanks :)\nThis isn't talked about much.</p>",
      "rawMarkdown": "Thanks :)\nThis isn't talked about much.",
      "votes": null
    },
    {
      "id": "301186",
      "postDate": "03/22/2018 13:00:52",
      "content": "<p>Thanks . Really helpful </p>",
      "rawMarkdown": "Thanks . Really helpful",
      "votes": null
    },
    {
      "id": "302376",
      "postDate": "03/24/2018 00:47:52",
      "content": "<p>Thank you!</p>",
      "rawMarkdown": "Thank you!",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 163788,
      "author_name": "toregil",
      "author_url": "",
      "post_date": "02/26/2017 00:01:17",
      "content": "<p>Thanks! But I assume the cartoon cats should be labelled 0.1.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 163957,
      "author_name": "evenoldridge",
      "author_url": "",
      "post_date": "02/27/2017 00:57:32",
      "content": "<p>Thanks.  That's really helpful.  I'd done a rudimentary attempt at this but had missed a number of these.  </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 163970,
      "author_name": "paulwsave",
      "author_url": "",
      "post_date": "02/27/2017 03:18:23",
      "content": "<p>Thanks Fabian! I started looking through the training data, but did not get though them yet... Very helpful!</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 165102,
      "author_name": "heryson",
      "author_url": "",
      "post_date": "03/03/2017 18:03:07",
      "content": "<p>Thanks for pointing it out, that's very fair !</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 169518,
      "author_name": "simongrest",
      "author_url": "",
      "post_date": "03/21/2017 09:31:27",
      "content": "<p>Thanks - very useful</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 182817,
      "author_name": "gumplus",
      "author_url": "",
      "post_date": "05/16/2017 01:01:34",
      "content": "<p>Thanks! Can you xplain how to find the wrong images via what method?</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 300914,
      "author_name": "svaisakh",
      "author_url": "",
      "post_date": "03/22/2018 03:56:12",
      "content": "<p>Thanks :)\nThis isn't talked about much.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 301186,
      "author_name": "archiies",
      "author_url": "",
      "post_date": "03/22/2018 13:00:52",
      "content": "<p>Thanks . Really helpful </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 302376,
      "author_name": "sunshinezhihuo",
      "author_url": "",
      "post_date": "03/24/2018 00:47:52",
      "content": "<p>Thank you!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "163291": "Hi,\n\nas already brought up in the previous iteration of this competition some of the training data is not exactly perfect. compare here:\nhttps://www.kaggle.com/c/dogs-vs-cats/discussion/5892\n\nI discovered some of these mistakes just now as I was looking through my models predictions on my cross validation set to see if I could find some systematic problem that would help me improve and get past @tunguz or @guangsha in the final days...\n\nin the spirit of fairness, here are the 52 relabels of the training data that I will now try to use to improve my model.\n\ndownload attachment or on github:\nhttps://gist.github.com/lefant/bb9c304cfcf5da0f9f652570cc54893a\n\ncheers and good luck,\n\nFabian",
    "163788": "Thanks! But I assume the cartoon cats should be labelled 0.1.",
    "163957": "Thanks.  That's really helpful.  I'd done a rudimentary attempt at this but had missed a number of these.",
    "163970": "Thanks Fabian! I started looking through the training data, but did not get though them yet... Very helpful!",
    "165102": "Thanks for pointing it out, that's very fair !",
    "169518": "Thanks - very useful",
    "182817": "Thanks! Can you xplain how to find the wrong images via what method?",
    "300914": "Thanks :)\nThis isn't talked about much.",
    "301186": "Thanks . Really helpful",
    "302376": "Thank you!"
  },
  "source": "meta"
}