{
  "id": 502476,
  "title": "Number of parameters required to pass baseline criteria",
  "url": "/competitions/leap-atmospheric-physics-ai-climsim/discussion/502476",
  "author_name": "Talha_Karim_TK",
  "post_date": "2024-05-13T16:25:08.901000",
  "votes": 2,
  "comment_count": 4,
  "views": 0,
  "content": "<p>I am trying to train a dense NN model, but the model doesn't seem to converge. I have tried different ways to have the convergence. One way that I couldn't because of my hardware limit is to build a large model.<br>\nCan anyone share how much parameters they are having for sequential dense model to pass the baseline criteria. Thanks</p>",
  "messages": [
    {
      "id": 2811192,
      "postDate": "2024-05-13T16:25:08.903Z",
      "content": "<p>I am trying to train a dense NN model, but the model doesn't seem to converge. I have tried different ways to have the convergence. One way that I couldn't because of my hardware limit is to build a large model.<br>\nCan anyone share how much parameters they are having for sequential dense model to pass the baseline criteria. Thanks</p>",
      "rawMarkdown": "I am trying to train a dense NN model, but the model doesn't seem to converge. I have tried different ways to have the convergence. One way that I couldn't because of my hardware limit is to build a large model.\nCan anyone share how much parameters they are having for sequential dense model to pass the baseline criteria. Thanks",
      "votes": 1
    },
    {
      "id": 2823969,
      "postDate": "2024-05-19T14:06:16.090Z",
      "content": "<p>I have got the same result of the baseline UNet with a 235k parameters CNN</p>",
      "rawMarkdown": "I have got the same result of the baseline UNet with a 235k parameters CNN"
    },
    {
      "id": 2812961,
      "postDate": "2024-05-14T13:53:46.170Z",
      "content": "<p>My LB result is 5.8M.<br>\nBut I'm sure it can be done with less. </p>",
      "rawMarkdown": "My LB result is 5.8M.\nBut I'm sure it can be done with less. "
    },
    {
      "id": 2813350,
      "postDate": "2024-05-14T17:40:42.560Z",
      "content": "<p>I'm not sure what is meant by baseline. Anyway, with about 500k parameters in NN, and with 10% of data (for train, validation and test), it gives public R2 = 0.52.<br>\nI think lots can be improved with how data is preprocessed (normalized). </p>\n<p>(I'm doing everything here on Kaggle notebooks)</p>",
      "rawMarkdown": "I'm not sure what is meant by baseline. Anyway, with about 500k parameters in NN, and with 10% of data (for train, validation and test), it gives public R2 = 0.52.\nI think lots can be improved with how data is preprocessed (normalized). \n\n(I'm doing everything here on Kaggle notebooks)",
      "isDeleted": true,
      "replies": [
        {
          "id": 2813481,
          "postDate": "2024-05-14T19:16:54.900Z",
          "content": "<p>By baseline I meant the score \"0.66765\". Don't actually know what it is. But thanks for the information. Looks like my problem was not using enough data for training.</p>",
          "rawMarkdown": "By baseline I meant the score \"0.66765\". Don't actually know what it is. But thanks for the information. Looks like my problem was not using enough data for training."
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 2823969,
      "author_name": "Amedeo Biolatti",
      "author_url": "",
      "post_date": "2024-05-19T14:06:16.090000",
      "content": "<p>I have got the same result of the baseline UNet with a 235k parameters CNN</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2812961,
      "author_name": "greySnow",
      "author_url": "",
      "post_date": "2024-05-14T13:53:46.170000",
      "content": "<p>My LB result is 5.8M.<br>\nBut I'm sure it can be done with less. </p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 2813350,
      "author_name": "",
      "author_url": "",
      "post_date": "2024-05-14T17:40:42.560000",
      "content": "<p>I'm not sure what is meant by baseline. Anyway, with about 500k parameters in NN, and with 10% of data (for train, validation and test), it gives public R2 = 0.52.<br>\nI think lots can be improved with how data is preprocessed (normalized). </p>\n<p>(I'm doing everything here on Kaggle notebooks)</p>",
      "votes": 0,
      "replies": [
        {
          "id": 2813481,
          "author_name": "Talha_Karim_TK",
          "author_url": "",
          "post_date": "2024-05-14T19:16:54.900000",
          "content": "<p>By baseline I meant the score \"0.66765\". Don't actually know what it is. But thanks for the information. Looks like my problem was not using enough data for training.</p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2811192": "I am trying to train a dense NN model, but the model doesn't seem to converge. I have tried different ways to have the convergence. One way that I couldn't because of my hardware limit is to build a large model.\nCan anyone share how much parameters they are having for sequential dense model to pass the baseline criteria. Thanks",
    "2823969": "I have got the same result of the baseline UNet with a 235k parameters CNN",
    "2812961": "My LB result is 5.8M.\nBut I'm sure it can be done with less. ",
    "2813350": "I'm not sure what is meant by baseline. Anyway, with about 500k parameters in NN, and with 10% of data (for train, validation and test), it gives public R2 = 0.52.\nI think lots can be improved with how data is preprocessed (normalized). \n\n(I'm doing everything here on Kaggle notebooks)"
  }
}