{
  "id": 533608,
  "title": "What does sigma_user for individual prediction mean?",
  "url": "/competitions/ariel-data-challenge-2024/discussion/533608",
  "author_name": "DennisSakva",
  "post_date": "2024-09-12T04:59:47.300000",
  "votes": 2,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hi, everyone</p>\n<p>Can someone explain what sigma_user (that one that we submit) for individual predictions/wavelength mean? </p>\n<p>Thanks!</p>",
  "messages": [
    {
      "id": 2986955,
      "postDate": "2024-09-12T08:05:44.500Z",
      "content": "<p>If you calculate the partial derivative of the loss function by sigma(planet,wavelength) you will find out that it zeros for sigma=y-mu. So, the best sigma is simply the difference between the target and the prediction. From here we can see the meaning: sigma is the prediction, of how far from the target is our prediction for mu.</p>",
      "rawMarkdown": "If you calculate the partial derivative of the loss function by sigma(planet,wavelength) you will find out that it zeros for sigma=y-mu. So, the best sigma is simply the difference between the target and the prediction. From here we can see the meaning: sigma is the prediction, of how far from the target is our prediction for mu.",
      "votes": 9,
      "replies": [
        {
          "id": 2987286,
          "postDate": "2024-09-12T14:59:44.653Z",
          "content": "<p>Thanks, greySnow! I think you are correct.</p>",
          "rawMarkdown": "Thanks, greySnow! I think you are correct."
        }
      ]
    },
    {
      "id": 2988052,
      "postDate": "2024-09-13T10:31:42.810Z",
      "content": "<p>For 777145011, I have normalized star flux uncertainty of 0.00739. This underestimates the (r/R)^2 uncertainty by a factor 4.4 (normlized Chi^2). There must be some earlier uncertainty that propagates; say, the residual from how well the chip readout fits a Fraunhofer (sinc^2) distribution… ?</p>",
      "rawMarkdown": "For 777145011, I have normalized star flux uncertainty of 0.00739. This underestimates the (r/R)^2 uncertainty by a factor 4.4 (normlized Chi^2). There must be some earlier uncertainty that propagates; say, the residual from how well the chip readout fits a Fraunhofer (sinc^2) distribution... ?",
      "votes": 1
    },
    {
      "id": 2986826,
      "postDate": "2024-09-12T04:59:47.300Z",
      "content": "<p>Hi, everyone</p>\n<p>Can someone explain what sigma_user (that one that we submit) for individual predictions/wavelength mean? </p>\n<p>Thanks!</p>",
      "rawMarkdown": "Hi, everyone\n\nCan someone explain what sigma_user (that one that we submit) for individual predictions/wavelength mean? \n\nThanks!",
      "votes": 2
    },
    {
      "id": 2987968,
      "postDate": "2024-09-13T08:26:47.100Z",
      "content": "<p>When we take the partial derivative of the loss function with respect to sigma (planet, wavelength), we find that it equals zero when sigma is the difference between the target and the prediction. Therefore, the best sigma is simply how much the target value differs from the predicted value. This shows that sigma represents the prediction of how far off our estimate is from the actual target.</p>",
      "rawMarkdown": "When we take the partial derivative of the loss function with respect to sigma (planet, wavelength), we find that it equals zero when sigma is the difference between the target and the prediction. Therefore, the best sigma is simply how much the target value differs from the predicted value. This shows that sigma represents the prediction of how far off our estimate is from the actual target.",
      "votes": -6,
      "replies": [
        {
          "id": 2988002,
          "postDate": "2024-09-13T09:29:42.817Z",
          "rawMarkdown": "",
          "votes": 1,
          "isDeleted": true
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2986955,
      "author_name": "greySnow",
      "author_url": "",
      "post_date": "2024-09-12T08:05:44.500000",
      "content": "<p>If you calculate the partial derivative of the loss function by sigma(planet,wavelength) you will find out that it zeros for sigma=y-mu. So, the best sigma is simply the difference between the target and the prediction. From here we can see the meaning: sigma is the prediction, of how far from the target is our prediction for mu.</p>",
      "votes": 9,
      "replies": [
        {
          "id": 2987286,
          "author_name": "DennisSakva",
          "author_url": "",
          "post_date": "2024-09-12T14:59:44.653000",
          "content": "<p>Thanks, greySnow! I think you are correct.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 2988052,
      "author_name": "Tord Malmgren",
      "author_url": "",
      "post_date": "2024-09-13T10:31:42.810000",
      "content": "<p>For 777145011, I have normalized star flux uncertainty of 0.00739. This underestimates the (r/R)^2 uncertainty by a factor 4.4 (normlized Chi^2). There must be some earlier uncertainty that propagates; say, the residual from how well the chip readout fits a Fraunhofer (sinc^2) distribution… ?</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 2987968,
      "author_name": "费文轩",
      "author_url": "",
      "post_date": "2024-09-13T08:26:47.100000",
      "content": "<p>When we take the partial derivative of the loss function with respect to sigma (planet, wavelength), we find that it equals zero when sigma is the difference between the target and the prediction. Therefore, the best sigma is simply how much the target value differs from the predicted value. This shows that sigma represents the prediction of how far off our estimate is from the actual target.</p>",
      "votes": -6,
      "replies": [
        {
          "id": 2988002,
          "author_name": "",
          "author_url": "",
          "post_date": "2024-09-13T09:29:42.817000",
          "content": "",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2986955": "If you calculate the partial derivative of the loss function by sigma(planet,wavelength) you will find out that it zeros for sigma=y-mu. So, the best sigma is simply the difference between the target and the prediction. From here we can see the meaning: sigma is the prediction, of how far from the target is our prediction for mu.",
    "2988052": "For 777145011, I have normalized star flux uncertainty of 0.00739. This underestimates the (r/R)^2 uncertainty by a factor 4.4 (normlized Chi^2). There must be some earlier uncertainty that propagates; say, the residual from how well the chip readout fits a Fraunhofer (sinc^2) distribution... ?",
    "2986826": "Hi, everyone\n\nCan someone explain what sigma_user (that one that we submit) for individual predictions/wavelength mean? \n\nThanks!",
    "2987968": "When we take the partial derivative of the loss function with respect to sigma (planet, wavelength), we find that it equals zero when sigma is the difference between the target and the prediction. Therefore, the best sigma is simply how much the target value differs from the predicted value. This shows that sigma represents the prediction of how far off our estimate is from the actual target."
  }
}