{
  "id": 541536,
  "title": "how to find sigma value to solve 0 score problem??",
  "url": "/competitions/ariel-data-challenge-2024/discussion/541536",
  "author_name": "",
  "post_date": "2024-10-20T04:48:19.029000",
  "votes": -4,
  "comment_count": 6,
  "views": 0,
  "content": "<p>in my case i find stranded deviation  between the wl_predicted and wl_true values does it is correct or not</p>",
  "messages": [
    {
      "id": 3023521,
      "postDate": "2024-10-20T16:43:18.007Z",
      "content": "<p>This is the biggest mistery for this competition. For the same prediction model, different approach of sigma estimation can give results from 0 to 0.56. My biggest puzzle is that my local test shows a better appraoch for sigma, but leadboard gives me 0 score! Very sad. My approach must be overfit badly.</p>",
      "rawMarkdown": "This is the biggest mistery for this competition. For the same prediction model, different approach of sigma estimation can give results from 0 to 0.56. My biggest puzzle is that my local test shows a better appraoch for sigma, but leadboard gives me 0 score! Very sad. My approach must be overfit badly.",
      "votes": 1,
      "replies": [
        {
          "id": 3024851,
          "postDate": "2024-10-22T04:47:56.123Z",
          "content": "<p>what kind of approach did you use</p>",
          "rawMarkdown": "what kind of approach did you use\n",
          "votes": 2,
          "replies": [
            {
              "id": 3025470,
              "postDate": "2024-10-22T19:49:08.817Z",
              "content": "<p>The constant sigma (like 0.000145 ) gave me safe results, but not good enough. If I tried to use std which varies by wavelength and planet, most of the time I got 0 score. </p>",
              "rawMarkdown": "The constant sigma (like 0.000145 ) gave me safe results, but not good enough. If I tried to use std which varies by wavelength and planet, most of the time I got 0 score. ",
              "votes": 1
            },
            {
              "id": 3025713,
              "postDate": "2024-10-23T05:09:15.940Z",
              "content": "<p>how about we train a model to find sigma values!! or else we can find a base sigma value for model from the training set and use it as test sigma values</p>",
              "rawMarkdown": "how about we train a model to find sigma values!! or else we can find a base sigma value for model from the training set and use it as test sigma values",
              "votes": 1
            },
            {
              "id": 3027921,
              "postDate": "2024-10-25T12:51:49.257Z",
              "content": "<p>the model can be severely overfitted on star0 and star1 data distributions</p>",
              "rawMarkdown": "the model can be severely overfitted on star0 and star1 data distributions",
              "votes": 2
            },
            {
              "id": 3028079,
              "postDate": "2024-10-25T15:44:03.890Z",
              "content": "<p>can you explain about star0 and star1 </p>",
              "rawMarkdown": "can you explain about star0 and star1 \n",
              "votes": 1
            }
          ]
        }
      ]
    },
    {
      "id": 3022916,
      "postDate": "2024-10-20T04:48:19.030Z",
      "content": "<p>in my case i find stranded deviation  between the wl_predicted and wl_true values does it is correct or not</p>",
      "rawMarkdown": "in my case i find stranded deviation  between the wl_predicted and wl_true values does it is correct or not",
      "votes": -4
    }
  ],
  "comments": [
    {
      "id": 3023521,
      "author_name": "John",
      "author_url": "",
      "post_date": "2024-10-20T16:43:18.007000",
      "content": "<p>This is the biggest mistery for this competition. For the same prediction model, different approach of sigma estimation can give results from 0 to 0.56. My biggest puzzle is that my local test shows a better appraoch for sigma, but leadboard gives me 0 score! Very sad. My approach must be overfit badly.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 3024851,
          "author_name": "",
          "author_url": "",
          "post_date": "2024-10-22T04:47:56.123000",
          "content": "<p>what kind of approach did you use</p>",
          "votes": 2,
          "replies": [
            {
              "id": 3025470,
              "author_name": "John",
              "author_url": "",
              "post_date": "2024-10-22T19:49:08.817000",
              "content": "<p>The constant sigma (like 0.000145 ) gave me safe results, but not good enough. If I tried to use std which varies by wavelength and planet, most of the time I got 0 score. </p>",
              "votes": 1,
              "replies": []
            },
            {
              "id": 3025713,
              "author_name": "",
              "author_url": "",
              "post_date": "2024-10-23T05:09:15.940000",
              "content": "<p>how about we train a model to find sigma values!! or else we can find a base sigma value for model from the training set and use it as test sigma values</p>",
              "votes": 1,
              "replies": []
            },
            {
              "id": 3027921,
              "author_name": "lllleeeo",
              "author_url": "",
              "post_date": "2024-10-25T12:51:49.257000",
              "content": "<p>the model can be severely overfitted on star0 and star1 data distributions</p>",
              "votes": 2,
              "replies": []
            },
            {
              "id": 3028079,
              "author_name": "",
              "author_url": "",
              "post_date": "2024-10-25T15:44:03.890000",
              "content": "<p>can you explain about star0 and star1 </p>",
              "votes": 1,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3023521": "This is the biggest mistery for this competition. For the same prediction model, different approach of sigma estimation can give results from 0 to 0.56. My biggest puzzle is that my local test shows a better appraoch for sigma, but leadboard gives me 0 score! Very sad. My approach must be overfit badly.",
    "3022916": "in my case i find stranded deviation  between the wl_predicted and wl_true values does it is correct or not"
  }
}