{
  "id": 206536,
  "title": "LB of single fold is 0.956，How can I get 0.960+？ ",
  "url": "/competitions/ranzcr-clip-catheter-line-classification/discussion/206536",
  "author_name": "",
  "post_date": "2020-12-25T06:15:20.226300500Z",
  "votes": null,
  "comment_count": 4,
  "views": 0,
  "content": "<p>Is there ways to get 0.960+？,thanks</p>",
  "messages": [
    {
      "id": "1125860",
      "postDate": "12/25/2020 06:15:20",
      "content": "<p>Is there ways to get 0.960+？,thanks</p>",
      "rawMarkdown": "Is there ways to get 0.960+？,thanks",
      "votes": null
    },
    {
      "id": "1126477",
      "postDate": "12/25/2020 16:34:26",
      "content": "<p>Ensembles, Augmentations, resizing scaling +training searching</p>",
      "rawMarkdown": "Ensembles, Augmentations, resizing scaling +training searching",
      "votes": null
    },
    {
      "id": "1128009",
      "postDate": "12/27/2020 05:21:32",
      "content": "<p>what model do you use?I only to get 0.950 use resnext50,and 5 fold.</p>",
      "rawMarkdown": "what model do you use?I only to get 0.950 use resnext50,and 5 fold.",
      "votes": null
    },
    {
      "id": "1128789",
      "postDate": "12/27/2020 18:33:18",
      "content": "<p>For the moment i just use a simble averaging of four CNN models, resnet, efficient, inception, Xception. While each one of these single models give around to 95-95.6, their averaging ensemble (pred1+pred2+pred3+pred4)/4 gave me a 96% score.</p>",
      "rawMarkdown": "For the moment i just use a simble averaging of four CNN models, resnet, efficient, inception, Xception. While each one of these single models give around to 95-95.6, their averaging ensemble (pred1+pred2+pred3+pred4)/4 gave me a 96% score.",
      "votes": null
    },
    {
      "id": "1128793",
      "postDate": "12/27/2020 18:35:53",
      "content": "<p>Now i want to try to take somehow advantage of the annotations too. They may give some information too in this ensemble.</p>",
      "rawMarkdown": "Now i want to try to take somehow advantage of the annotations too. They may give some information too in this ensemble.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1126477,
      "author_name": "",
      "author_url": "",
      "post_date": "12/25/2020 16:34:26",
      "content": "<p>Ensembles, Augmentations, resizing scaling +training searching</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1128009,
      "author_name": "bcwang",
      "author_url": "",
      "post_date": "12/27/2020 05:21:32",
      "content": "<p>what model do you use?I only to get 0.950 use resnext50,and 5 fold.</p>",
      "votes": null,
      "replies": [
        {
          "id": 1128789,
          "author_name": "",
          "author_url": "",
          "post_date": "12/27/2020 18:33:18",
          "content": "<p>For the moment i just use a simble averaging of four CNN models, resnet, efficient, inception, Xception. While each one of these single models give around to 95-95.6, their averaging ensemble (pred1+pred2+pred3+pred4)/4 gave me a 96% score.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1128793,
          "author_name": "",
          "author_url": "",
          "post_date": "12/27/2020 18:35:53",
          "content": "<p>Now i want to try to take somehow advantage of the annotations too. They may give some information too in this ensemble.</p>",
          "votes": null,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "1125860": "Is there ways to get 0.960+？,thanks",
    "1126477": "Ensembles, Augmentations, resizing scaling +training searching",
    "1128009": "what model do you use?I only to get 0.950 use resnext50,and 5 fold.",
    "1128789": "For the moment i just use a simble averaging of four CNN models, resnet, efficient, inception, Xception. While each one of these single models give around to 95-95.6, their averaging ensemble (pred1+pred2+pred3+pred4)/4 gave me a 96% score.",
    "1128793": "Now i want to try to take somehow advantage of the annotations too. They may give some information too in this ensemble."
  },
  "source": "meta"
}