{
  "id": 216941,
  "title": "How to explain this case?",
  "url": "/competitions/hubmap-kidney-segmentation/discussion/216941",
  "author_name": "",
  "post_date": "2021-02-04T15:46:55.317163800Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>I have trained the same parameter with effectUnet two times (single model no kfold). but with one LB: 0.836 and the other is 0.846. how could it shake so much(0.01) ?  <br>\njust put your opinions on the comment field.<br>\nThank you :)</p>",
  "messages": [
    {
      "id": "1186127",
      "postDate": "02/04/2021 15:46:55",
      "content": "<p>I have trained the same parameter with effectUnet two times (single model no kfold). but with one LB: 0.836 and the other is 0.846. how could it shake so much(0.01) ?  <br>\njust put your opinions on the comment field.<br>\nThank you :)</p>",
      "rawMarkdown": "I have trained the same parameter with effectUnet two times (single model no kfold). but with one LB: 0.836 and the other is 0.846. how could it shake so much(0.01) ?  \njust put your opinions on the comment field.\nThank you :)",
      "votes": null
    },
    {
      "id": "1186563",
      "postDate": "02/04/2021 22:34:49",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/puyuzhou\" target=\"_blank\">@puyuzhou</a> … the randomness of deeplearning models explains a lot of it.<br>\nI train my models with stratified KFold and it can have the same variations as you mention on different runs. </p>",
      "rawMarkdown": "Hi @puyuzhou ... the randomness of deeplearning models explains a lot of it.\nI train my models with stratified KFold and it can have the same variations as you mention on different runs.",
      "votes": null
    },
    {
      "id": "1191485",
      "postDate": "02/08/2021 13:40:14",
      "content": "<p>If you want the same results on different runs you should fix the seed numbers.</p>",
      "rawMarkdown": "If you want the same results on different runs you should fix the seed numbers.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 1186563,
      "author_name": "rsmits",
      "author_url": "",
      "post_date": "02/04/2021 22:34:49",
      "content": "<p>Hi <a href=\"https://www.kaggle.com/puyuzhou\" target=\"_blank\">@puyuzhou</a> … the randomness of deeplearning models explains a lot of it.<br>\nI train my models with stratified KFold and it can have the same variations as you mention on different runs. </p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 1191485,
      "author_name": "erikdali",
      "author_url": "",
      "post_date": "02/08/2021 13:40:14",
      "content": "<p>If you want the same results on different runs you should fix the seed numbers.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "1186127": "I have trained the same parameter with effectUnet two times (single model no kfold). but with one LB: 0.836 and the other is 0.846. how could it shake so much(0.01) ?  \njust put your opinions on the comment field.\nThank you :)",
    "1186563": "Hi @puyuzhou ... the randomness of deeplearning models explains a lot of it.\nI train my models with stratified KFold and it can have the same variations as you mention on different runs.",
    "1191485": "If you want the same results on different runs you should fix the seed numbers."
  },
  "source": "meta"
}