{
  "id": 410911,
  "title": "Some Questions during Trainings ><",
  "url": "/competitions/google-research-identify-contrails-reduce-global-warming/discussion/410911",
  "author_name": "chusheng chen",
  "post_date": "2023-05-17T00:30:21.141000",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Hi , dear all senior kagglers and grand masters…. i am new in kaggle <br>\nand i face a problem during training ….<br>\nthat is i found that the performance of my model doesn't improve after some training ….<br>\nwhat can i do to improve and solve it…?<br>\nmy training strategy :</p>\n<ol>\n<li>filter out all the mark with only background , to reduce training time…</li>\n<li>patch training input images into serveral patches</li>\n<li>training with segmentation model (UNet )</li>\n<li>focalloss is used as loss function</li>\n</ol>\n<p>any senior can give me some advices?thanks…..</p>",
  "messages": [
    {
      "id": 2262454,
      "postDate": "2023-05-17T00:31:14.230Z",
      "content": "<p>because of it spends too much time for one epoch and seem doesn't improve .. so i interrupt the training….</p>",
      "rawMarkdown": "because of it spends too much time for one epoch and seem doesn't improve .. so i interrupt the training....",
      "votes": 1
    },
    {
      "id": 2262453,
      "postDate": "2023-05-17T00:30:21.143Z",
      "content": "<p>Hi , dear all senior kagglers and grand masters…. i am new in kaggle <br>\nand i face a problem during training ….<br>\nthat is i found that the performance of my model doesn't improve after some training ….<br>\nwhat can i do to improve and solve it…?<br>\nmy training strategy :</p>\n<ol>\n<li>filter out all the mark with only background , to reduce training time…</li>\n<li>patch training input images into serveral patches</li>\n<li>training with segmentation model (UNet )</li>\n<li>focalloss is used as loss function</li>\n</ol>\n<p>any senior can give me some advices?thanks…..</p>",
      "rawMarkdown": "Hi , dear all senior kagglers and grand masters.... i am new in kaggle \nand i face a problem during training ....\nthat is i found that the performance of my model doesn't improve after some training ....\nwhat can i do to improve and solve it...?\nmy training strategy :\n1. filter out all the mark with only background , to reduce training time...\n2. patch training input images into serveral patches\n3. training with segmentation model (UNet )\n4. focalloss is used as loss function\n\nany senior can give me some advices?thanks.....",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 2262454,
      "author_name": "chusheng chen",
      "author_url": "",
      "post_date": "2023-05-17T00:31:14.230000",
      "content": "<p>because of it spends too much time for one epoch and seem doesn't improve .. so i interrupt the training….</p>",
      "votes": 1,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "2262454": "because of it spends too much time for one epoch and seem doesn't improve .. so i interrupt the training....",
    "2262453": "Hi , dear all senior kagglers and grand masters.... i am new in kaggle \nand i face a problem during training ....\nthat is i found that the performance of my model doesn't improve after some training ....\nwhat can i do to improve and solve it...?\nmy training strategy :\n1. filter out all the mark with only background , to reduce training time...\n2. patch training input images into serveral patches\n3. training with segmentation model (UNet )\n4. focalloss is used as loss function\n\nany senior can give me some advices?thanks....."
  }
}