{
  "id": 497782,
  "title": "Cosine Weighting?",
  "url": "/competitions/leap-atmospheric-physics-ai-climsim/discussion/497782",
  "author_name": "Sebastian Hoffmann",
  "post_date": "2024-04-25T17:15:07.893000",
  "votes": 5,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hey,</p>\n<p>Quick question. Given that the dataset is sampled from lat-lon grids, will the evaluation take the oversampling of the poles into account (e.g. by cosine weighting) or has this already been taken into account during dataset creation via subsampling?</p>\n<p>If not, this is something which should be addressed ASAP, because otherwise you'll end up with models which are highly overfitted to the polar regions.</p>",
  "messages": [
    {
      "id": 2775500,
      "postDate": "2024-04-25T17:15:07.893Z",
      "content": "<p>Hey,</p>\n<p>Quick question. Given that the dataset is sampled from lat-lon grids, will the evaluation take the oversampling of the poles into account (e.g. by cosine weighting) or has this already been taken into account during dataset creation via subsampling?</p>\n<p>If not, this is something which should be addressed ASAP, because otherwise you'll end up with models which are highly overfitted to the polar regions.</p>",
      "rawMarkdown": "Hey,\n\nQuick question. Given that the dataset is sampled from lat-lon grids, will the evaluation take the oversampling of the poles into account (e.g. by cosine weighting) or has this already been taken into account during dataset creation via subsampling?\n\nIf not, this is something which should be addressed ASAP, because otherwise you'll end up with models which are highly overfitted to the polar regions.",
      "votes": 5
    },
    {
      "id": 2775936,
      "postDate": "2024-04-25T22:24:37.563Z",
      "content": "<p>Cosine-weighting is not needed since the data isn't sampled from a lat-lon grid. You can check the grid information here:<br>\n<a href=\"https://huggingface.co/datasets/LEAP/ClimSim_low-res/blob/main/ClimSim_low-res_grid-info.nc\" target=\"_blank\">https://huggingface.co/datasets/LEAP/ClimSim_low-res/blob/main/ClimSim_low-res_grid-info.nc</a></p>\n<p>I also created a plot of the grid points:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14891093%2F4afa50959ed2e9fcd4e843ae7b2bb5a3%2Fgrid_points.png?generation=1714083822484969&amp;alt=media\"></p>",
      "rawMarkdown": "Cosine-weighting is not needed since the data isn't sampled from a lat-lon grid. You can check the grid information here:\nhttps://huggingface.co/datasets/LEAP/ClimSim_low-res/blob/main/ClimSim_low-res_grid-info.nc\n\nI also created a plot of the grid points:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14891093%2F4afa50959ed2e9fcd4e843ae7b2bb5a3%2Fgrid_points.png?generation=1714083822484969&alt=media)",
      "votes": 6,
      "replies": [
        {
          "id": 2782765,
          "postDate": "2024-04-29T12:50:58.747Z",
          "content": "<p>Hey Felix,</p>\n<p>Thank a lot! This clears things up a lot!</p>",
          "rawMarkdown": "Hey Felix,\n\nThank a lot! This clears things up a lot!",
          "votes": 1
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2775936,
      "author_name": "Felix Yang",
      "author_url": "",
      "post_date": "2024-04-25T22:24:37.563000",
      "content": "<p>Cosine-weighting is not needed since the data isn't sampled from a lat-lon grid. You can check the grid information here:<br>\n<a href=\"https://huggingface.co/datasets/LEAP/ClimSim_low-res/blob/main/ClimSim_low-res_grid-info.nc\" target=\"_blank\">https://huggingface.co/datasets/LEAP/ClimSim_low-res/blob/main/ClimSim_low-res_grid-info.nc</a></p>\n<p>I also created a plot of the grid points:<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14891093%2F4afa50959ed2e9fcd4e843ae7b2bb5a3%2Fgrid_points.png?generation=1714083822484969&amp;alt=media\"></p>",
      "votes": 6,
      "replies": [
        {
          "id": 2782765,
          "author_name": "Sebastian Hoffmann",
          "author_url": "",
          "post_date": "2024-04-29T12:50:58.747000",
          "content": "<p>Hey Felix,</p>\n<p>Thank a lot! This clears things up a lot!</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2775500": "Hey,\n\nQuick question. Given that the dataset is sampled from lat-lon grids, will the evaluation take the oversampling of the poles into account (e.g. by cosine weighting) or has this already been taken into account during dataset creation via subsampling?\n\nIf not, this is something which should be addressed ASAP, because otherwise you'll end up with models which are highly overfitted to the polar regions.",
    "2775936": "Cosine-weighting is not needed since the data isn't sampled from a lat-lon grid. You can check the grid information here:\nhttps://huggingface.co/datasets/LEAP/ClimSim_low-res/blob/main/ClimSim_low-res_grid-info.nc\n\nI also created a plot of the grid points:\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F14891093%2F4afa50959ed2e9fcd4e843ae7b2bb5a3%2Fgrid_points.png?generation=1714083822484969&alt=media)"
  }
}