{
  "id": 237997,
  "title": "helpful methods",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/237997",
  "author_name": "",
  "post_date": "2021-05-11T00:18:01.735717900Z",
  "votes": 7,
  "comment_count": 7,
  "views": 0,
  "content": "<p>We got 0.949 in our best final submissions(chose wrong, they are very similar). </p>\n<p>One and only one of the key methods is to use average and weighted average to lower the  danger of the  label in d488.</p>\n<p>here is my paras:<br>\nclass config:<br>\nbackbone = 'efficientnet-b7'<br>\nACTIVATION = 'sigmoid'<br>\nENCODER_WEIGHTS = 'imagenet'<br>\nlr= 0.0002<br>\nepochs= 50<br>\nbatch_size= 15<br>\nnum_workers=4<br>\ndropout = 0.7<br>\nsize = 512<br>\nshifted = True<br>\nexternal = True<br>\npatience = 8<br>\npatience_rate = 0.5</p>",
  "messages": [
    {
      "id": "1301087",
      "postDate": "05/11/2021 00:18:01",
      "content": "<p>We got 0.949 in our best final submissions(chose wrong, they are very similar). </p>\n<p>One and only one of the key methods is to use average and weighted average to lower the  danger of the  label in d488.</p>\n<p>here is my paras:<br>\nclass config:<br>\nbackbone = 'efficientnet-b7'<br>\nACTIVATION = 'sigmoid'<br>\nENCODER_WEIGHTS = 'imagenet'<br>\nlr= 0.0002<br>\nepochs= 50<br>\nbatch_size= 15<br>\nnum_workers=4<br>\ndropout = 0.7<br>\nsize = 512<br>\nshifted = True<br>\nexternal = True<br>\npatience = 8<br>\npatience_rate = 0.5</p>",
      "rawMarkdown": "We got 0.949 in our best final submissions(chose wrong, they are very similar). \n\nOne and only one of the key methods is to use average and weighted average to lower the  danger of the  label in d488.\n\nhere is my paras:\nclass config:\nbackbone = 'efficientnet-b7'\nACTIVATION = 'sigmoid'\nENCODER_WEIGHTS = 'imagenet'\nlr= 0.0002\nepochs= 50\nbatch_size= 15\nnum_workers=4\ndropout = 0.7\nsize = 512\nshifted = True\nexternal = True\npatience = 8\npatience_rate = 0.5",
      "votes": null
    },
    {
      "id": "1301862",
      "postDate": "05/11/2021 09:25:06",
      "content": "<p>what does <code>mean and weighted mean</code> meaning ？</p>",
      "rawMarkdown": "what does `mean and weighted mean` meaning ？",
      "votes": null
    },
    {
      "id": "1301929",
      "postDate": "05/11/2021 10:13:39",
      "content": "<p>equals to average and weighted average. revised.</p>",
      "rawMarkdown": "equals to average and weighted average. revised.",
      "votes": null
    },
    {
      "id": "1302023",
      "postDate": "05/11/2021 11:10:18",
      "content": "<p>Thanks for sharing, Congrats!</p>",
      "rawMarkdown": "Thanks for sharing, Congrats!",
      "votes": null
    },
    {
      "id": "1302036",
      "postDate": "05/11/2021 11:15:01",
      "content": "<p>each pix minus std and divide by mean rather than all divide 255?</p>",
      "rawMarkdown": "each pix minus std and divide by mean rather than all divide 255?",
      "votes": null
    },
    {
      "id": "1302041",
      "postDate": "05/11/2021 11:16:21",
      "content": "<p>BTW, did you use handlabel for d48 during training in final submittion?</p>",
      "rawMarkdown": "BTW, did you use handlabel for d48 during training in final submittion?",
      "votes": null
    },
    {
      "id": "1302446",
      "postDate": "05/11/2021 14:48:51",
      "content": "<p>sure I used them. and I tuned its weight very low.</p>",
      "rawMarkdown": "sure I used them. and I tuned its weight very low.",
      "votes": null
    },
    {
      "id": "1302705",
      "postDate": "05/11/2021 17:09:54",
      "content": "<p>Charlie has done weighted average on models without d488 and I use normal average base on his results with models with d488 .Thus, the weight of d488 is small but not loss its effect.</p>",
      "rawMarkdown": "Charlie has done weighted average on models without d488 and I use normal average base on his results with models with d488 .Thus, the weight of d488 is small but not loss its effect.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1301862,
      "author_name": "drzhuzhe",
      "author_url": "",
      "post_date": "05/11/2021 09:25:06",
      "content": "<p>what does <code>mean and weighted mean</code> meaning ？</p>",
      "votes": null,
      "replies": [
        {
          "id": 1301929,
          "author_name": "southsakura",
          "author_url": "",
          "post_date": "05/11/2021 10:13:39",
          "content": "<p>equals to average and weighted average. revised.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1302036,
          "author_name": "drzhuzhe",
          "author_url": "",
          "post_date": "05/11/2021 11:15:01",
          "content": "<p>each pix minus std and divide by mean rather than all divide 255?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1302041,
          "author_name": "drzhuzhe",
          "author_url": "",
          "post_date": "05/11/2021 11:16:21",
          "content": "<p>BTW, did you use handlabel for d48 during training in final submittion?</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1302446,
          "author_name": "puyuzhou",
          "author_url": "",
          "post_date": "05/11/2021 14:48:51",
          "content": "<p>sure I used them. and I tuned its weight very low.</p>",
          "votes": null,
          "replies": []
        },
        {
          "id": 1302705,
          "author_name": "southsakura",
          "author_url": "",
          "post_date": "05/11/2021 17:09:54",
          "content": "<p>Charlie has done weighted average on models without d488 and I use normal average base on his results with models with d488 .Thus, the weight of d488 is small but not loss its effect.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 1302023,
      "author_name": "salimkhazem",
      "author_url": "",
      "post_date": "05/11/2021 11:10:18",
      "content": "<p>Thanks for sharing, Congrats!</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1301087": "We got 0.949 in our best final submissions(chose wrong, they are very similar). \n\nOne and only one of the key methods is to use average and weighted average to lower the  danger of the  label in d488.\n\nhere is my paras:\nclass config:\nbackbone = 'efficientnet-b7'\nACTIVATION = 'sigmoid'\nENCODER_WEIGHTS = 'imagenet'\nlr= 0.0002\nepochs= 50\nbatch_size= 15\nnum_workers=4\ndropout = 0.7\nsize = 512\nshifted = True\nexternal = True\npatience = 8\npatience_rate = 0.5",
    "1301862": "what does `mean and weighted mean` meaning ？",
    "1301929": "equals to average and weighted average. revised.",
    "1302023": "Thanks for sharing, Congrats!",
    "1302036": "each pix minus std and divide by mean rather than all divide 255?",
    "1302041": "BTW, did you use handlabel for d48 during training in final submittion?",
    "1302446": "sure I used them. and I tuned its weight very low.",
    "1302705": "Charlie has done weighted average on models without d488 and I use normal average base on his results with models with d488 .Thus, the weight of d488 is small but not loss its effect."
  },
  "source": "meta"
}