{
  "id": 50420,
  "title": "Error of data ?",
  "url": "/competitions/dog-breed-identification/discussion/50420",
  "author_name": "",
  "post_date": "2018-02-24T00:26:52.311868400Z",
  "votes": null,
  "comment_count": 5,
  "views": 0,
  "content": "<p>I found that there are some identical pictures in training set and test set, such as 03e1d2ee5fd90aef036c90a9e7f81177.jpg in training set and 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test set, but their labels are not the same. I submitted the corresponding labels and found the log loss on LB increase. Am I wrong?</p>",
  "messages": [
    {
      "id": "287233",
      "postDate": "02/24/2018 00:26:52",
      "content": "<p>I found that there are some identical pictures in training set and test set, such as 03e1d2ee5fd90aef036c90a9e7f81177.jpg in training set and 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test set, but their labels are not the same. I submitted the corresponding labels and found the log loss on LB increase. Am I wrong?</p>",
      "rawMarkdown": "I found that there are some identical pictures in training set and test set, such as 03e1d2ee5fd90aef036c90a9e7f81177.jpg in training set and 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test set, but their labels are not the same. I submitted the corresponding labels and found the log loss on LB increase. Am I wrong?",
      "votes": null
    },
    {
      "id": "287885",
      "postDate": "02/26/2018 06:30:53",
      "content": "<p>What method did you use to compare them?</p>",
      "rawMarkdown": "What method did you use to compare them?",
      "votes": null
    },
    {
      "id": "287894",
      "postDate": "02/26/2018 06:45:23",
      "content": "<p>For example, in label.csv, the label of 03e1d2ee5fd90aef036c90a9e7f81177.jpg is bernese_mountain_dog. I found that 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test is the same with that image. So in submission file, I made the predicted probability of bernese_mountain_dog of this image is 1 or 0.99405 and  the predicted probability of other breeds are 0 or 0.00005. Then I submitted and found the score in LB increase.</p>",
      "rawMarkdown": "For example, in label.csv, the label of 03e1d2ee5fd90aef036c90a9e7f81177.jpg is bernese_mountain_dog. I found that 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test is the same with that image. So in submission file, I made the predicted probability of bernese_mountain_dog of this image is 1 or 0.99405 and  the predicted probability of other breeds are 0 or 0.00005. Then I submitted and found the score in LB increase.",
      "votes": null
    },
    {
      "id": "288275",
      "postDate": "02/27/2018 04:45:36",
      "content": "<p>is it possible to get log loss less then 0.3 using kaggle dataset without stanford dataset</p>",
      "rawMarkdown": "is it possible to get log loss less then 0.3 using kaggle dataset without stanford dataset",
      "votes": null
    },
    {
      "id": "288277",
      "postDate": "02/27/2018 04:49:18",
      "content": "<p>yes, i have seen that in other forums</p>",
      "rawMarkdown": "yes, i have seen that in other forums",
      "votes": null
    },
    {
      "id": "288308",
      "postDate": "02/27/2018 07:16:54",
      "content": "<p>can you point out some forums</p>",
      "rawMarkdown": "can you point out some forums",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 287885,
      "author_name": "sharwon",
      "author_url": "",
      "post_date": "02/26/2018 06:30:53",
      "content": "<p>What method did you use to compare them?</p>",
      "votes": null,
      "replies": [
        {
          "id": 287894,
          "author_name": "shiqitao",
          "author_url": "",
          "post_date": "02/26/2018 06:45:23",
          "content": "<p>For example, in label.csv, the label of 03e1d2ee5fd90aef036c90a9e7f81177.jpg is bernese_mountain_dog. I found that 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test is the same with that image. So in submission file, I made the predicted probability of bernese_mountain_dog of this image is 1 or 0.99405 and  the predicted probability of other breeds are 0 or 0.00005. Then I submitted and found the score in LB increase.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 288275,
      "author_name": "sainathreddy",
      "author_url": "",
      "post_date": "02/27/2018 04:45:36",
      "content": "<p>is it possible to get log loss less then 0.3 using kaggle dataset without stanford dataset</p>",
      "votes": null,
      "replies": [
        {
          "id": 288277,
          "author_name": "pukkinming",
          "author_url": "",
          "post_date": "02/27/2018 04:49:18",
          "content": "<p>yes, i have seen that in other forums</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 288308,
          "author_name": "sainathreddy",
          "author_url": "",
          "post_date": "02/27/2018 07:16:54",
          "content": "<p>can you point out some forums</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "287233": "I found that there are some identical pictures in training set and test set, such as 03e1d2ee5fd90aef036c90a9e7f81177.jpg in training set and 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test set, but their labels are not the same. I submitted the corresponding labels and found the log loss on LB increase. Am I wrong?",
    "287885": "What method did you use to compare them?",
    "287894": "For example, in label.csv, the label of 03e1d2ee5fd90aef036c90a9e7f81177.jpg is bernese_mountain_dog. I found that 25c88ab100f3e86e4a0215bfbcafd2c4.jpg in test is the same with that image. So in submission file, I made the predicted probability of bernese_mountain_dog of this image is 1 or 0.99405 and  the predicted probability of other breeds are 0 or 0.00005. Then I submitted and found the score in LB increase.",
    "288275": "is it possible to get log loss less then 0.3 using kaggle dataset without stanford dataset",
    "288277": "yes, i have seen that in other forums",
    "288308": "can you point out some forums"
  },
  "source": "meta"
}