{
  "id": 86567,
  "title": "People use dirty tricks should be removed",
  "url": "/competitions/histopathologic-cancer-detection/discussion/86567",
  "author_name": "zxyu",
  "post_date": "2019-03-25T05:17:07.289000",
  "votes": 4,
  "comment_count": 3,
  "views": 0,
  "content": "<p>kaggle should removed those who use dirty tricks. and  forbid them to take part in Competition for several months even years. </p>",
  "messages": [
    {
      "id": 499694,
      "postDate": "2019-03-25T05:17:07.290Z",
      "content": "<p>kaggle should removed those who use dirty tricks. and  forbid them to take part in Competition for several months even years. </p>",
      "rawMarkdown": "kaggle should removed those who use dirty tricks. and  forbid them to take part in Competition for several months even years. ",
      "votes": 4
    },
    {
      "id": 499751,
      "postDate": "2019-03-25T06:21:27.653Z",
      "content": "<p>Hello, let me share my story and you can than remove me if any illegal. I have LB 1.000. I understand now how to do this is different ways.\n- In an earlier competition we used external data and we were top 1% and postcompetition we have leading result, so I know it is worth using external data.\n- Organizers provided link to external data (<a href=\"https://github.com/basveeling/pcam\">https://github.com/basveeling/pcam</a>)\n- I had .98% before using it with a combination of different resnet and inception architectures, that was a week ago top 10 place.\n- I inserted than the pcam test and validation data and my score improved surprisingly.\n- Was not difficult to figure out what is going on. I am not a hash creation person but image analyst and a double for loop (probably taking slightly longer than hash matching) that finds correlation between images makes the job.\n- Unfortunately this competition is over if there is no new test set provided within days as anybody can make now arbitrary scores (I can make you .9899) \n- Even if there was a new test set with more clever naming, if it is from pcam, it will take 1 hour to fish it out.</p>\n\n<p>Anyways luckily there is no money involved, as I think we all learned a lot.</p>",
      "rawMarkdown": "Hello, let me share my story and you can than remove me if any illegal. I have LB 1.000. I understand now how to do this is different ways.\n- In an earlier competition we used external data and we were top 1% and postcompetition we have leading result, so I know it is worth using external data.\n- Organizers provided link to external data (https://github.com/basveeling/pcam)\n- I had .98% before using it with a combination of different resnet and inception architectures, that was a week ago top 10 place.\n- I inserted than the pcam test and validation data and my score improved surprisingly.\n- Was not difficult to figure out what is going on. I am not a hash creation person but image analyst and a double for loop (probably taking slightly longer than hash matching) that finds correlation between images makes the job.\n- Unfortunately this competition is over if there is no new test set provided within days as anybody can make now arbitrary scores (I can make you .9899) \n- Even if there was a new test set with more clever naming, if it is from pcam, it will take 1 hour to fish it out.\n\nAnyways luckily there is no money involved, as I think we all learned a lot.",
      "votes": 4
    },
    {
      "id": 504210,
      "postDate": "2019-03-31T06:40:27.040Z",
      "content": "<p>agree lol</p>",
      "rawMarkdown": "agree lol"
    },
    {
      "id": 499736,
      "postDate": "2019-03-25T06:11:50.017Z",
      "rawMarkdown": "",
      "votes": 1,
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 499751,
      "author_name": "Peter Horvath",
      "author_url": "",
      "post_date": "2019-03-25T06:21:27.653000",
      "content": "<p>Hello, let me share my story and you can than remove me if any illegal. I have LB 1.000. I understand now how to do this is different ways.\n- In an earlier competition we used external data and we were top 1% and postcompetition we have leading result, so I know it is worth using external data.\n- Organizers provided link to external data (<a href=\"https://github.com/basveeling/pcam\">https://github.com/basveeling/pcam</a>)\n- I had .98% before using it with a combination of different resnet and inception architectures, that was a week ago top 10 place.\n- I inserted than the pcam test and validation data and my score improved surprisingly.\n- Was not difficult to figure out what is going on. I am not a hash creation person but image analyst and a double for loop (probably taking slightly longer than hash matching) that finds correlation between images makes the job.\n- Unfortunately this competition is over if there is no new test set provided within days as anybody can make now arbitrary scores (I can make you .9899) \n- Even if there was a new test set with more clever naming, if it is from pcam, it will take 1 hour to fish it out.</p>\n\n<p>Anyways luckily there is no money involved, as I think we all learned a lot.</p>",
      "votes": 4,
      "replies": []
    },
    {
      "id": 504210,
      "author_name": "keeplearning",
      "author_url": "",
      "post_date": "2019-03-31T06:40:27.040000",
      "content": "<p>agree lol</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 499736,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-03-25T06:11:50.017000",
      "content": "",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "499694": "kaggle should removed those who use dirty tricks. and  forbid them to take part in Competition for several months even years. ",
    "499751": "Hello, let me share my story and you can than remove me if any illegal. I have LB 1.000. I understand now how to do this is different ways.\n- In an earlier competition we used external data and we were top 1% and postcompetition we have leading result, so I know it is worth using external data.\n- Organizers provided link to external data (https://github.com/basveeling/pcam)\n- I had .98% before using it with a combination of different resnet and inception architectures, that was a week ago top 10 place.\n- I inserted than the pcam test and validation data and my score improved surprisingly.\n- Was not difficult to figure out what is going on. I am not a hash creation person but image analyst and a double for loop (probably taking slightly longer than hash matching) that finds correlation between images makes the job.\n- Unfortunately this competition is over if there is no new test set provided within days as anybody can make now arbitrary scores (I can make you .9899) \n- Even if there was a new test set with more clever naming, if it is from pcam, it will take 1 hour to fish it out.\n\nAnyways luckily there is no money involved, as I think we all learned a lot.",
    "504210": "agree lol",
    "499736": ""
  }
}