{
  "id": 85149,
  "title": "Accuracy is about 0.856, bug aug score is about 0.94",
  "url": "/competitions/histopathologic-cancer-detection/discussion/85149",
  "author_name": "",
  "post_date": "2019-03-22T01:20:11.452933300Z",
  "votes": null,
  "comment_count": 7,
  "views": 0,
  "content": "<p>I have trained the  deep learning model and get a funny result. The Accuracy is much lower than the aug score and the train's scores is lower than the validation's score! So does any one have some idea about it?</p>",
  "messages": [
    {
      "id": "496210",
      "postDate": "03/22/2019 01:20:11",
      "content": "<p>I have trained the  deep learning model and get a funny result. The Accuracy is much lower than the aug score and the train's scores is lower than the validation's score! So does any one have some idea about it?</p>",
      "rawMarkdown": "I have trained the  deep learning model and get a funny result. The Accuracy is much lower than the aug score and the train's scores is lower than the validation's score! So does any one have some idea about it?",
      "votes": null
    },
    {
      "id": "496216",
      "postDate": "03/22/2019 01:25:57",
      "content": "<p>You are most likely using the hard score instead of a soft score. LB here uses a soft score. If you change probability (like 0.8 to 1, 0.01 to 0) to hard encoded integers, your score will drop about 0.1.</p>",
      "rawMarkdown": "You are most likely using the hard score instead of a soft score. LB here uses a soft score. If you change probability (like 0.8 to 1, 0.01 to 0) to hard encoded integers, your score will drop about 0.1.",
      "votes": null
    },
    {
      "id": "496238",
      "postDate": "03/22/2019 01:54:42",
      "content": "<p>@Hanke Chen Sorry but I don't exactly know what the LB means</p>",
      "rawMarkdown": "Hanke Chen Sorry but I don't exactly know what the LB means",
      "votes": null
    },
    {
      "id": "496245",
      "postDate": "03/22/2019 02:08:08",
      "content": "<p>Hahaha, this is also the first question when joined Kaggle. LB stands for <code>leaderboard</code>. It usually means the score you got from the <strong>public</strong> leaderboard.</p>",
      "rawMarkdown": "Hahaha, this is also the first question when joined Kaggle. LB stands for `leaderboard`. It usually means the score you got from the **public** leaderboard.",
      "votes": null
    },
    {
      "id": "496266",
      "postDate": "03/22/2019 02:35:27",
      "content": "<p>@Hanke Chen\nif your meas were that the predictions all in{0,1], My answer is yes</p>",
      "rawMarkdown": "Hanke Chen\nif your meas were that the predictions all in{0,1], My answer is yes",
      "votes": null
    },
    {
      "id": "496268",
      "postDate": "03/22/2019 02:35:57",
      "content": "<p>@HankeChen</p>",
      "rawMarkdown": "HankeChen",
      "votes": null
    },
    {
      "id": "496272",
      "postDate": "03/22/2019 02:41:26",
      "content": "<p><code>Hanke Chen</code> is my nickname. You should @KokeCacao. Don't use predictions in 0s or 1s. Use decimal probability.</p>",
      "rawMarkdown": "`Hanke Chen` is my nickname. You should @KokeCacao. Don't use predictions in 0s or 1s. Use decimal probability.",
      "votes": null
    },
    {
      "id": "496283",
      "postDate": "03/22/2019 02:54:48",
      "content": "<p>@KokeCacao I'm sorry for that.  </p>",
      "rawMarkdown": "KokeCacao I'm sorry for that.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 496216,
      "author_name": "kokecacao",
      "author_url": "",
      "post_date": "03/22/2019 01:25:57",
      "content": "<p>You are most likely using the hard score instead of a soft score. LB here uses a soft score. If you change probability (like 0.8 to 1, 0.01 to 0) to hard encoded integers, your score will drop about 0.1.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 496238,
      "author_name": "muxinghan",
      "author_url": "",
      "post_date": "03/22/2019 01:54:42",
      "content": "<p>@Hanke Chen Sorry but I don't exactly know what the LB means</p>",
      "votes": null,
      "replies": [
        {
          "id": 496245,
          "author_name": "kokecacao",
          "author_url": "",
          "post_date": "03/22/2019 02:08:08",
          "content": "<p>Hahaha, this is also the first question when joined Kaggle. LB stands for <code>leaderboard</code>. It usually means the score you got from the <strong>public</strong> leaderboard.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 496266,
      "author_name": "muxinghan",
      "author_url": "",
      "post_date": "03/22/2019 02:35:27",
      "content": "<p>@Hanke Chen\nif your meas were that the predictions all in{0,1], My answer is yes</p>",
      "votes": null,
      "replies": [
        {
          "id": 496272,
          "author_name": "kokecacao",
          "author_url": "",
          "post_date": "03/22/2019 02:41:26",
          "content": "<p><code>Hanke Chen</code> is my nickname. You should @KokeCacao. Don't use predictions in 0s or 1s. Use decimal probability.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 496268,
      "author_name": "muxinghan",
      "author_url": "",
      "post_date": "03/22/2019 02:35:57",
      "content": "<p>@HankeChen</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 496283,
      "author_name": "muxinghan",
      "author_url": "",
      "post_date": "03/22/2019 02:54:48",
      "content": "<p>@KokeCacao I'm sorry for that.  </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "496210": "I have trained the  deep learning model and get a funny result. The Accuracy is much lower than the aug score and the train's scores is lower than the validation's score! So does any one have some idea about it?",
    "496216": "You are most likely using the hard score instead of a soft score. LB here uses a soft score. If you change probability (like 0.8 to 1, 0.01 to 0) to hard encoded integers, your score will drop about 0.1.",
    "496238": "Hanke Chen Sorry but I don't exactly know what the LB means",
    "496245": "Hahaha, this is also the first question when joined Kaggle. LB stands for `leaderboard`. It usually means the score you got from the **public** leaderboard.",
    "496266": "Hanke Chen\nif your meas were that the predictions all in{0,1], My answer is yes",
    "496268": "HankeChen",
    "496272": "`Hanke Chen` is my nickname. You should @KokeCacao. Don't use predictions in 0s or 1s. Use decimal probability.",
    "496283": "KokeCacao I'm sorry for that."
  },
  "source": "meta"
}